JeVois Tech Zone - Recent questions and answers in Programmer Questions

How to solve this Error: ModuleNotFoundError: No module named 'torch._C'

Sat, 14 Aug 2021 09:50:30 +0000

Hello. I have the following problem:

In my Python Software I use : from detecto import core, utils, visualize

After this I imediately receive this error:

ERR Log::warnAndIgnoreException: Caught exception from the Python interpreter:

ERR Log::warnAndIgnoreException: Python traceback (most recent calls last):

ERR Log::warnAndIgnoreException: File /jevois/modules/Dambrowsky/PythonTest/PythonTest.py

ERR Log::warnAndIgnoreException: Line 166, in process:

ERR Log::warnAndIgnoreException: File /usr/lib/python3.7/site-packages/detecto/core.py

ERR Log::warnAndIgnoreException: Line 4, in <module>:

ERR Log::warnAndIgnoreException: File /usr/lib/python3.7/site-packages/torch/__init__.py

ERR Log::warnAndIgnoreException: Line 79, in <module>:

ERR Log::warnAndIgnoreException: ModuleNotFoundError: No module named 'torch._C'

I have installed the following Libraries:

hell pip list

Package Version

--------------------- ---------------

arrow 0.10.0

asn1crypto 0.24.0

async-timeout 3.0.1

attrs 18.2.0

backcall 0.1.0

beautifulsoup4 4.5.3

bitstring 3.1.5

cached-property 1.4.0

cbor 1.0.0

certifi 2018.11.29

cffi 1.11.2

characteristic 14.3.0

chardet 3.0.4

constantly 15.1.0

cryptography 2.3.1

cycler 0.10.0

daemonize 2.4.7

DataProperty 0.29.1

decorator 4.0.11

detecto 1.2.1

humanize 0.5.1

idna 2.8

iowait 0.2

ipython-genutils 0.2.0

json-schema-validator 2.4.1

jsonmodels 2.4

jsonschema 2.5.1

kiwisolver 1.3.1

Logbook 1.4.3

lxml 4.3.3

markdown2 2.3.6

matplotlib 3.4.2

mbstrdecoder 0.2.2

more-itertools 4.0.1

msgpack 0.6.1

multidict 4.5.2

numpy 1.16.3

pandas 1.1.0

path.py 10.5

pathtools 0.1.2

pathvalidate 0.16.0

pexpect 4.6.0

pickleshare 0.7.4

Pillow 8.3.0

pip 19.1

ply 3.11

posix-ipc 1.0.4

protobuf 3.6.0

psutil 5.6.2

ptyprocess 0.5.1

py 1.8.0

py-ubjson 0.12.0

pycparser 2.19

pycrypto 2.6.1

pyinotify 0.9.6

pylibdmtx 0.1.9

pylru 1.0.9

pyparsing 3.0.0b2

pytablereader 0.15.0

python-dateutil 2.8.0

PyTrie 0.3.1

pytz 2018.9

PyYAML 4.2b4

requests 2.21.0

scandir 1.9.0

schedule 0.6.0

scipy 1.3.0

see 1.4.1

semver 2.8.1

setuptools 41.0.1

sh 1.12.14

simplegeneric 0.8.1

simplejson 3.11.1

SimpleSQLite 0.15.0

six 1.12.0

sortedcontainers 2.1.0

tabledata 0.0.5

tempora 1.7

torch 1.1.0

torchvision 0.3.0a0+bbd363c

tqdm 4.61.2

traitlets 4.3.2

typepy 0.0.20

ujson 1.35

urllib3 1.24.2

uvloop 0.12.2

validators 0.12.2

visitor 0.1.3

Whoosh 2.7.4

xlrd 1.0.0

It looks like the Library torch._C is not available. Is there anybody who can help me?

Thank you very much.

Peter Dambrowsky

How to uninstall a Library from the Linux partition of the microSD

Tue, 10 Aug 2021 07:26:15 +0000

When I have installed a Library with the Inventor using a .whl file (it is functioning only with the inventor under Ubuntu) and I want do delete it later, how do I have to do it. Is there a special shell command? I tried to do it directly under usr/lib/python3.7/site-packages on the microSD Linux partition. But here I could see, that I have not the rights to delete something.

An answer would realy help me.

Best regards

Peter

Answered: Problem when installing customized module to jevois

Tue, 09 Mar 2021 08:48:06 +0000

I have no idea about this but in this situation you get help with expert and resolve your problems.

Answered: How to create an own MobileNetSSD caffe model

Sat, 09 Jan 2021 09:12:42 +0000

Up ! Artificial Intelligence is hard !

Answered: By now, is it possible to read IMU raw data using python?

Sat, 09 Jan 2021 09:11:58 +0000

Yes it's possible but, go show tutorial, python is hard to use !

Answered: please moderate this message so I can post.

Fri, 28 Aug 2020 21:39:10 +0000

just found this now buried among 600 spam post attempts... I believe you need one more approved one before you get out of moderation.

Answered: support for data matrix / small 2D barcode

Wed, 24 Jun 2020 22:40:17 +0000

I remember working on this a while back but I am not sure I ever converged... Can you please check this:

http://jevois.org/moddoc/PyDMTX/modinfo.html

with a recent JeVois version (like 1.14.0). The fact that there is no screenshot means maybe it is not fully working yet. Basically we need to install libdmtx and pydmtx on the platform hardware. As far as I remember, it is quite slow, so maybe that is why we did not complete it.

Answered: Using tensorflow in python

Wed, 24 Jun 2020 22:34:45 +0000

This is not supported, though you should be able to run your model using Python OpenCV, see for example

http://jevois.org/moddoc/PyClassificationDNN/modinfo.html

http://jevois.org/moddoc/PyDetectionDNN/modinfo.html

Unfortunately cross-compiling TF is becoming increasingly difficult because of their bazel build system, but we will look into whether we could compile the python bindings into our TF module.

Answered: Headless jevois : frame to Serial

Wed, 24 Jun 2020 22:31:12 +0000

In no-USB on, you also need to issue a "streamon" command once your module is configured and ready to start work. Before that, it is just waiting idle. Please see

http://jevois.org/tutorials/ProgrammerPythonSaveImages.html

for example.

Answered: Using tensorflow within a python module

Wed, 24 Jun 2020 22:27:52 +0000

We do not support TF directly in Python on JeVois, only C++ for now. However, maybe using OpenCV would help (you should be able to run TF models using the OpenCV framework). That one is fully accessible from Python.

Answered: Is it possible to send both image and IMU data back to host computer through USB cable?

Wed, 27 May 2020 09:33:08 +0000

I had the same problem, I was searching for a better way to send the IMU data without the coupling of the image data. I watched a lot of youtube tutorials, but I didn't found anything that could help to answer this question, but take a lot to this webpage: https://www.techradar.com/reviews/nordvpn, maybe you will find something. I didn't try yet, but my friend recommended to search there.

Answered: Not able to install developer software the quick way as given in jevois.usc.edu

Tue, 12 May 2020 07:47:57 +0000

yes, so sorry, the certificate expired and I had to go to the server physically today to renew it. Please try again, it should work now.

Answered: Possible software error in the JeVois cam A33

Wed, 15 Apr 2020 11:25:46 +0000

I found a solution to send date and time from an Arduino Mega about some seconds after power on (the red led has to be on)

You have to do it with this Arduino software:

#define JeVois Serial2 // for example com2

  JeVois.begin(115200);     //Verbindung mit JeVois-Kamera
  JeVois.setTimeout(1000000);

  JeVois.println("date " + DatumZeit);

And now very important:

In the variable DatumZeit has to be the actual date and time with a string like this "date 041112302020.00" it means "date MMDDhhmmCCYY.ss" M = Month, D = Day, h = hour, m = minute, CC = the first 2 digits of the year, YY = the last 2 digits of the year and .ss = seconds (I use here always 00).

Only this way functions perfect.

Best regards

Peter Dambrowsky

Answered: How to get Serial data from jevois to NodeMCU ?

Fri, 03 Apr 2020 14:21:21 +0000

It seems that it's not possible to use both RxTx pins and serial monitor. It works using following code:

#include <ESP8266WiFi.h>
#include <ESP8266httpUpdate.h>
#define INLEN 256

char instr[INLEN + 1];

const char* ssid = "SSID";
const char* password = "PWD";
const int led=2;

void setup() {
Serial.begin(115200);
WiFi.mode(WIFI_STA);
WiFi.begin(ssid, password);
while (WiFi.waitForConnectResult() != WL_CONNECTED) {
delay(1000);
ESP.restart();
}
}

void loop() {
digitalWrite(led, LOW);
Serial.println("info");

String val = "test=";

while (Serial.available()) {
val.concat((char)Serial.read());
}

HTTPClient http; //Declare object of class HTTPClient
http.begin("http://192.168.1.33/duino/read_post_request.php");
http.addHeader("Content-Type", "application/x-www-form-urlencoded");
int httpCode = http.POST(val);
http.end();
digitalWrite(led, HIGH);
delay(1000);
}

Answered: Object detection

Wed, 11 Mar 2020 20:58:02 +0000

The ObjectDetect module http://jevois.org/moddoc/ObjectDetect/modinfo.html has several parameters you can tune if your objects are not detected with the default settings. The default settings assume that your object has unique texture elements, and will not work well with objects that are very simple, e.g., a white square object.

First try higher "distthresh". This is how similar keypoints must be between learned image and current image.

Then try to broaden the range for "goodpts", this is how many matching key points are needed between learned image and current image.

Then try to lower "hessian" which controls how sharp keypoints have two be to be considered.

Answered: How can I store a Video frame by frame to the sdcard of my A33 cam

Wed, 11 Mar 2020 20:51:19 +0000

Looks like you are opening "cap" as a video capture device but that is not available in JeVois modules.

The core JeVois software does the capture for you. All you need is to write a process() function that will be called each time a new frame is available from the sensor. In that function you can save your frames. Make sure you make "out" a class data member of your module so that it persists across multiple calls to process()

Have a look here for some guidance:

http://jevois.org/tutorials/ProgrammerInvHello.html

http://jevois.org/doc/ModulePythonTutorial.html

Start with the code for Hello. Open "out" (renamed self.out) in __init__ of the module). Then in process() add your "out.write(frame)". I recommend that you use an absolute path for your saved video (e.g., /jevois/myvideo.avi)

Answered: DemoSaliency headless

Wed, 11 Mar 2020 20:43:04 +0000

You need to create your own module for that, or add headless capability and whiteness computation to ours.

Starting from http://jevois.org/basedoc/jvpkg_2modules_2JeVois_2DemoSaliency_2DemoSaliency_8C_source.html

you would copy the whole process() function into a new version that does not have a second arg for the output frame. This is the function you need for headless operation. Then delete everything related to drawing the output composite display. Then add your whiteness evaluation code.

Here are a few pointers to get started:

http://jevois.org/tutorials/ProgrammerSetup.html

http://jevois.org/doc/ModuleTutorial.html

http://jevois.org/tutorials/UserHeadless.html

Answered: How to detect new objects in Jevois inventor

Wed, 11 Mar 2020 20:34:20 +0000

Oh, it should be either in the "Console" tab of JeVois Inventor (easiest), or in a terminal connected to JeVois over serial port.

Does the jevois-inventor provides the feature to flash the current time off host computer

Fri, 06 Mar 2020 17:09:45 +0000

Hello, I am looking for a python program to take the date and time from the host computer and send it to the Jevois A33 cam ( like I can do it in the console with the comand "date 2020.03.06-15:30:00".

I want to do this with a program line in Python on ervery start of the Cam.

Please help me.

Regards

Peter

Answered: Update the time in jevois each time automatically.

Mon, 02 Mar 2020 14:33:03 +0000

Please tell me more how the jevois-inventor can flash the current time from the host computer.

Thank you.

Regards Peter

Answered: How do I save JeVois camera images on my arduino

Fri, 14 Feb 2020 00:38:23 +0000

Have a look at http://jevois.org/qa/index.php?qa=3879

Answered: How do I save pictures if it's hooked up to a microcontroller

Fri, 14 Feb 2020 00:37:06 +0000

Try our SaveVideo module, and see here for a tutorial:

http://jevois.org/moddoc/SaveVideo/modinfo.html

http://jevois.org/tutorials/User120fps.html

That will save on the microSD inside JeVois.

Answered: Read Only File System after sendSerial?

Tue, 11 Feb 2020 21:52:54 +0000

Can you use an absolute path when you open the file, like "/jevois/targetData.txt"?

It might be that for some strange reason you do not always end up being in the module's directory as you should be (maybe something funny going on during instantiation of a python module, the constructor may be called before we change directory to the module's directory). The /jevois partition should be writable.

Or, in the inventor console, try

shell dmesg

and look for messages that would signal some disk error and re-mounting read-only. Or try

shell touch /jevois/hello

and then

shell ls /jevois

and see if you get errors and whether the file was created.

If that does not help, we have had strange situations also (but hard to reproduce) when trying to save python code to SD repeatedly and across multiple days. I think it might have to do with date and pycache. And maybe a broken pycache triggers the remount to read-only. So I would recommend the following: after JeVois boots up, fire up the inventor and go to the System tab and set the date/time on the camera to that of your host. Then you should have no problem editing, etc. My guess about this is that since by default jevois starts on Jan 1, 1970 (since there is no battery for the clock), depending on when you edit and try to save, you may end up with saving the source python file earlier than you did last time you booted up JeVois, and so now you end up with the compiled python code in pycache being in the future compared to the time at which you save your source python file. That may upset python.

Answered: Jevois-daemon freezing after ~2 seconds

Tue, 11 Feb 2020 21:23:14 +0000

Yes, we have seen similar things using virtual box on MacOS just recently.

Could you please try to shut down your virtual box completely (to make sure it releases the USB ports) and then get the native MacOS version of JeVois Inventor (or OBS Studio or Webcamoid) and see whether that works? In our testing it did work. So we concluded something was going on with virtual box, maybe the video drivers or such. We updated to the latest virtual box and latest extension pack and rebooted the host (yes, this is needed, even on Mac), and the problem was gone somehow...

Answered: How to fix "Are you using old TFLite binary with newer model?"

Fri, 31 Jan 2020 02:22:14 +0000

Darn, we are kind of stuck at TF 1.9.9, please see the exact version in

https://github.com/jevois/jevoisbase/blob/master/Contrib/reinstall.sh

we get this exact version (the long number is the commit ID on GitHub):

get_github tensorflow tensorflow 25c197e02393bd44f50079945409009dd4d434f8

And then we compile only the C++ source files we need (discarding anything that would try to open a window and such), and those get integrated into libjevjoisbase, please see here, section about TensorFlow:

https://github.com/jevois/jevoisbase/blob/master/CMakeLists.txt

We have tried to update to a newer version but their build system has changed and we have not yet been able to track down what exactly gets compiled and how through their bazel-based build. If you can somehow figure out an updated list of sources for the one that starts at line 378 of the above CMakeLists.txt, then we can try that update to a newer tensorflow again.

Otherwise, maybe you can make your model work under TF 1.9.9?

Answered: Sending commands from python to serial command line

Fri, 31 Jan 2020 02:13:21 +0000

This is not supported. Just to give you the historical background: we started with a C++ core framework. In there, all one can access, from within a vision module, all core functionality (like list the mappings, etc) by calling the appropriate functions on the JeVois Engine object (which is always at the root of a vision pipeline). Then we added the command-line interface as a way for the host computer to also access some of the core functions, over a serial port. Finally we added python support on top of the C++ core, with some direct bindings from the core to the Python side of things, shown here:

http://jevois.org/doc/ProgrammerPython.html

(see section on Python bindings provided by jevois core).

Now we did not provide a python binding for something like listmappings because a vision module (the python code you are writing) should not need to know that. It should work just for whatever mappings it can support, and the host computer is in charge of picking a mapping (and hence a vision module) it wants to run. If listmappings is indeed what you want, then can you tell us why you would need it? From within a vision module, you should never have to worry about that list. If you want something else, maybe we can add a python binding for that?

In the meantime, the best way to quickly solve your problem is to have your host computer issue some commands to the jevois command-line interface, for example like here:

http://jevois.org/tutorials/UserPanTilt.html

if you scroll down to the first piece of source code, line 43 the Arduino sends this to the JeVois camera:

SERIAL.println("setpar serlog None");

to configure it.

Or in this one:

http://jevois.org/tutorials/UserColorTracking.html

some python program running on the host sends some commands to JeVois as the sliders in the Python GUI are moved around.

Answered: Issue in running custom Tensorflow network

Tue, 28 Jan 2020 21:58:55 +0000

VGG-16 may indeed be too big for JeVois.

Can you email us your model files and params.cfg to jevois.org@gmail.com? We can try to load it here with a USB-to-serial diagnostic cable plugged in and we can tell you if the camera runs out of memory.

Answered: Tensorflow lite Object Detection

Tue, 28 Jan 2020 21:49:45 +0000

Looks like you have 4 output arrays, but TensorFlowEasy only works with 1 output. So you would have to create your own JeVois module to handle your output format. The easiest is to use OpenCV and Python. Skip the conversion to TFlite for now as I am not sure how well that works with OpenCV. Just use your baseline trained TF network. Then look at these two:

For recognition only (identify what is in the image): jevois module PyClassifcationDNN

http://jevois.org/basedoc/jvpkg_2modules_2JeVois_2PyClassificationDNN_2PyClassificationDNN_8py_source.html

For detection + identification (find bounding boxes around objects and identify the objects): jevois module PyDetectionDNN

http://jevois.org/basedoc/jvpkg_2modules_2JeVois_2PyDetectionDNN_2PyDetectionDNN_8py_source.html

You can edit those modules in JeVois Inventor and adapt them to your particular output format.

Answered: Any OpenCV Python demo to stream data directly from JeVois as a UVC camera?

Tue, 28 Jan 2020 18:50:43 +0000

yes, there is no separate python-libjevois, both the C++ code and the python bindings are in libjevois.so

as far as a demo to behave like a dumb camera, start with the Python Sandbox module:

http://jevois.org/basedoc/jvpkg_2modules_2JeVois_2PythonSandbox_2PythonSandbox_8py_source.html

and just replace line 62 with:

outimg = inimg

which will simply copy the image from the sensor (inning) to the one that is sent over USB to the host computer (outimg).

Answered: I've added a custom C++ module and JeVois Inventor can't connect to the camera

Tue, 28 Jan 2020 18:43:57 +0000

hum, that is very strange. Can you email us more info about your setup? I am guessing you have an entry for your module in videomappings.cfg already? One possibility is that on some machines (MacOS, Windows), the camera drivers get confused when trying to select among several video modes that only differ by their frame rate. This is not a problem when using guvcview, webcamoid, etc but it is a problem when using the Qt camera stack, which we use in the inventor. So maybe disable in videomappings.cfg any other modes that have the same output resolution but different frame rates.

Answered: I need a program in Phyton to integrate some more pictures.

Fri, 10 Jan 2020 18:33:06 +0000

You need to do that offline using whichever method is recommended for that particular network. Once it is retrained, you can copy the weight files to your microSD and edit the config file for your module to load that network by default.

For mobilenet + SSD Caffe model, you would usually use this module:

http://jevois.org/moddoc/DetectionDNN/modinfo.html

http://jevois.org/moddoc/PyDetectionDNN/modinfo.html

For the latter (Python version), you can just edit the Python code in JeVois inventor, add a new block like this one:

elif (model == 'MobileNetV2SSD'):

classnames = '/jevois/share/darknet/yolo/data/coco.names'

modelname = '/jevois/share/opencv-dnn/detection/ssd_mobilenet_v2_coco_2018_03_29.pb'

configname = '/jevois/share/opencv-dnn/detection/ssd_mobilenet_v2_coco_2018_03_29.pbtxt'

with the correct locations for the 3 files: List of class names, weight file, model file.

Answered: Headless mode not sending serial output for Python modules.

Tue, 03 Dec 2019 15:59:58 +0000

http://jevois.org/tutorials/UserHeadless.html Provides the basic outline of how to operate in the headless mode. (use setmapping2 instead of setmapping. setmapping2 is meant specifically for going headless).
One special case when you may not get any serial messages from jevois is if you try to send non-string characters, as in, if you send an array which does not contain ascii values corresponding to alphabets of numbers (It didn't work for me in C++ so I see no reason why it would work in python).

Answered: using one module inside another

Wed, 27 Nov 2019 04:52:13 +0000

Definitely, yes this is very nice to have. JeVois supports this, just the vocabulary is a bit different:

- you can create Component objects that may host a number of Parameters and any number of sub-components.

- Module derives from Component but is intended to be a terminal entity.

- in JeVois, you load and run one module at a time. The root of the hierarchy is the Engine, which holds the module as a sub-component, which in turn can hold more sub-components.

For more info, see

http://jevois.org/doc/classjevois_1_1Component.html

A good example is the ArUcoBlob module which has an ArUco subcomponent plus several BlobDetector sub-components. For more examples search for "addSubComponent" in the source tree of jevoisbase.

http://jevois.org/basedoc/jvpkg_2modules_2JeVois_2ArUcoBlob_2ArUcoBlob_8C_source.html

This one may also be a good starting point for you:

http://jevois.org/basedoc/classBufferedVideoReader.html

Answered: libjevoisebase components in your project

Wed, 27 Nov 2019 04:39:20 +0000

Sorry for the slow reply as your post was buried in a surge of spam posts we had to moderate. After a couple approved posts you will not be moderated anymore.

Yes, you do need to link against libjevjoisbase. Have a look at CMakeLists.txt that comes with samplemodule:

https://github.com/jevois/samplemodule/blob/master/CMakeLists.txt

you need to uncomment a few things to allow linking against libjevoisbase:

## If your module will use components provided by jevoisbase, uncomment the lines below:
if (JEVOIS_PLATFORM)
link_directories("/var/lib/jevois-microsd/lib/JeVois") # find libjevoisbase.so for platform

endif (JEVOIS_PLATFORM)
include_directories("/var/lib/jevois-build/usr/include") # find jevoisbase includes

target_link_libraries(SampleModule ${JEVOIS_OPENCV_LIBS} jevoisbase opencv_imgproc opencv_core)

Answered: Best module to use for high FPS vehicle & object detection in a road setting?

Thu, 24 Oct 2019 00:18:50 +0000

The main limitation of python is that it does not support true multithreading. So you are kind of stuck with using one core of JeVois when you could use 4 in C++ easily using std::async() to launch threads.

Now, having said that, you should also be aware that not much in terms of machine vision is actually implemented in python. Even OpenCV, that is all C++ under the hood. When you use OpenCV in python, that just calls the underlying C++ OpenCV implementation. For this reason, some OpenCV functions may still be highly efficient and multithreaded. So, usually, for pure OpenCV code, there is not much speed difference between python and C++, since python just calls the OpenCV C++ code. You would see big differences if, for example, you have a loop over all pixels in an image, and then do something in that loop. The loop may run much slower in python than C++. But as long as you only use image-level functions (like, for example, add two images) then the speed difference is negligible since the same optimized C++ code for matrix addition will be executed in both cases.

For TensorFlow and 83fps: this is with the smallest variant of MobileNets V1 quantized, that network runs at about 83fps. If you can cast your problem into something that a CNN could solve, then you can train your own mobilenet and run it on JeVois. See here:

http://jevois.org/tutorials/UserTensorFlowTraining.html

Answered: Convert module to “Supports mappings with NO USB output: YES”

Mon, 23 Sep 2019 16:15:31 +0000

yes, that can be done quite easily. You need to be setup for C++ programming, please see:

http://jevois.org/tutorials/ProgrammerVbox.html

http://jevois.org/tutorials/ProgrammerSetup.html

Then you need to add a member function

void process(jevois::RawImage const & input) override

{

...

}

to the DemoSaliency class in DemoSaliency.C

This function will be called by the JeVois core software when running in headless mode. To write the function, you take the other process() function and copy that code, then delete everything that involves the output image (outframe, outimg, etc).

Looking at the code at

http://jevois.org/basedoc/jvpkg_2modules_2JeVois_2DemoSaliency_2DemoSaliency_8C_source.html

it also looks like you would not need to run the saliency computation in a thread (which is parallelized with waiting for the out frame and drawing in it), you can just run

itsSaliency->process(inimg, true);

directly in the main thread of your process() function. I believe you then also should remove the call to

itsSaliency->waitUntilDoneWithInput();

as this is intended for use when itsSaliency->process() is running in a thread.

Have a look at this tutorial for a general idea of the various steps:

http://jevois.org/tutorials/ProgrammerSurprise.html

Answered: 1.13.0/1.14.0 - Simple JeVois-For-FRC module freezes camera quickly:-(

Tue, 17 Sep 2019 16:29:39 +0000

That error means JeVois was disconnected somehow. It could either be a crash, or a mechanical issue at the connector. USB connectors are not designed for mechanical stress, so you should use strain relief for the connector and cable when mounting on a robot. Have a look at this one, it is super nice:

https://www.thingiverse.com/thing:2734559

See how there is a clamp for the USB connector behind the camera. You secure the cable to the mount using a zip tie once the cable and camera are mounted.

On the software side, we have cameras running 24 hours a day for months here with no problem... In the JeVois inventor console or your Arduino console, try things like "shell free" and monitor the amount of free RAM over time. If it is going down (beyond small fluctuations), then maybe your code is leaking memory or crashing somehow. To investigate that, try to run our TensorFlowEasy code on your camera. This one is pretty brutal, gets near 400% of CPU usage. If that one is stable, then the hardware is ok. If that one crashes and you are not moving the camera around, maybe the hardware has a problem.

that video of a very short run also suggests that maybe you have a power issue. Make sure you use a Y cable, and that your computer can indeed source 500mA each port. Small boards like the raspberry pi typically cannot, since they are powered from a single USB port themselves.Have a look here for tips:

http://jevois.org/doc/UserConnect.html

Answered: Playback a video through Jevois?

Thu, 12 Sep 2019 16:11:45 +0000

Yes, this is supported, exactly for the purposes you mention (optimizing an algo).

When we implemented this, we assumed people would want to use this with jevois-daemon running on the host computer. Parameter cameradev selects the device for the camera, but that could also be a movie file:

 --cameradev (string) default=[/dev/video0]
    Camera device name (if starting with /dev/v...), or movie file name (e.g., movie.mpg) or image sequence (e.g., im%02d.jpg, to read frames im00.jpg, im01.jpg, etc).

from http://jevois.org/doc/UserCli.html

If running jevois-daemon on host (this requires that you install the jevois deb packages on an ubuntu computer), you would start it as follows:

jevois-daemon --cameradev=movie.mpg --videomapping=23 # select the mapping that corresponds to your module; use listmappings in the console to know that number.

If you are using JeVois Inventor and want to run the movie in your camera instead, put your movie in the JEVOIS partition of the SD card, and then, add this to JEVOIS:/config/params.cfg

cameradev=/jevois/movie.mpg

(note: this config file is for experts only and thus is not available in JeVois inventor. It is the system-global params.cfg as opposed to being each module's params.cfg, which is exposed in JeVois inventor. So you need to connect the SD card to your computer, or export it from a running jevois camera it using the button under the System tab of jevois inventor).

If you copy a file movie.mpg into the root of the JEVOIS partition, that is seen by the software running in the camera as /jevois/movie.mpg

Note: when we read a movie file, we automatically rescale it to whatever input resolution your module expects.

Note2: When the movie ends, we try to rewind it, but for some video formats this is not possible. In such case, your module may stall at the end of the movie with an error that it cannot read the next video frame.

Answered: headless not sending serial messages

Tue, 10 Sep 2019 16:46:14 +0000

oh, in setmapping2, since you will go headless, you only specify the camera sensor resolution, and you have no video output resolution.

So you should use 320 240 for your resolution (camera sensor resolution), not 320 264 (which would be an output resolution to stream to USB). The sensor does not support 320x264 capture and hence your setmapping2 probably fails on boot up.

I am guessing you have already seen http://jevois.org/tutorials/UserHeadless.html

Now for PyClassificationDNN, indeed we did not write processNoUSB(), so you would need to write it by just taking process() and stripping out anything that has to do with drawing an output image.

Answered: Add a new aruco image

Fri, 06 Sep 2019 16:10:49 +0000

Good question! With SURF key points (ObjectDetect or SaliencySurf), the problem will be that your shape is not very unique in the world, and hence you may get lots of false positives.

For ArUco, in part the algorithm is robust because of the more complex and carefully crafted shapes in the dictionary. So you may not get the same level of robustness with your shape. I don't think ArUco supports custom shapes, but I believe ARtoolkit (which we also support) does. However, you may still have issues with false alarms with that simple shape.

I wonder whether this one would be a good starting point: http://jevois.org/moddoc/FirstPython/modinfo.html

also see https://www.youtube.com/watch?v=BaWostkMClA

In there, we try to detect a U shape. If you edit the code in JeVois Inventor, have a look at lines 160 and later. This is where we enforce that the shape should look like a U:

- first, its convex hull should be quadrilateral

- then its area should be within some range (you can remove that test if you want)

- then (lines 179-180), the shape contour approximated as a polygon should have 8 vertices (replace that by 4)

- finally (lines 183-186), we check that the approximate polygon matches the fully detailed shape quite well. You could remove that first, then play with it later.

To understand this code, you need to look at the OpenCV docs:

https://docs.opencv.org/4.1.1/dd/d49/tutorial_py_contour_features.html

Also you may want to look online for "OpenCV shape detection python", and see, eg,

https://www.pyimagesearch.com/2016/02/08/opencv-shape-detection/

Answered: Cross-compiled module using BlobDetector : undefined symbol ?

Wed, 21 Aug 2019 19:46:38 +0000

did you start from the jevois sample module? If so, you need to edit your CMakeLists.txt to link against the jevoisbase library, since BlobDetector is implemented in that library.

There are some explanations in the CMakeLists.txt, please see https://github.com/jevois/samplemodule/blob/master/CMakeLists.txt

typically, you would need to uncomment the commands link_directories(), include_directories(), and add jevoisbase as a dependent library in the list in target_link_libraries()

If this does not work, please email us your source code (the whole module directory) and we will help you.

Answered: Communication With Multiple JeVois Cameras on a Raspberry Pi

Thu, 01 Aug 2019 16:47:27 +0000

I would definitely go with serial-over-USB for all cameras for this. You will need a USB hub that can power all the cameras, so that means it needs to have strong external power. For example, we use these with success here:

https://www.orico.shop/en/orico-7-port-usb-30-hub-aluminum-5gbps.html

You want to make sure the hub has enough power for what you need. For example, the one above states it has a 12V/2.5A power supply so that is 30 Watts going in. Each camera could use up to 3.8 Watts, say 4 Watts to allow for some margin. Also we should derate the 30 Watts a bit to account for 12V to 5V conversion, so say 24 Watts available to the USB ports, enough for 6 JeVois cameras (should be ok with 7 as the derating used here is quite severe).

Then use USB cables with 24AWG or lower power lines (see http://jevois.org/doc/UserConnect.html for details).

Finally, each camera will appear on the Pi as a USB-to-Serial adapter. This will create devices /dev/ttyACM0, /dev/ttyACM1, etc on your Pi.You should configure each camera to run in headless mode, see http://jevois.org/tutorials/UserHeadless.html

You can open each device in your application code running on the Pi, and read the data from it. Something similar to this: http://jevois.org/tutorials/UserParseSerial.html but you need to change the code so that it is non-blocking. That is, you just check if some new bytes are available on each port and accumulate the bytes into a buffer until you get a \n that indicates end of line. Once you have a complete line, you can parse that message. If you are programming in C++ see function readSome() here: http://jevois.org/doc/Serial_8C_source.html

Note for completeness that by default each JeVois camera requests the full USB video bandwidth for video streaming. This is to optimize latency of video. Video bandwidth is separate from serial bandwidth, so this does not affect the serial-over-USB ports. But if you wanted to stream video from several cameras simultaneously (as opposed to just serial messages like you want here), see http://jevois.org/doc/Multicam.html

Finally, the USB hub may enumerate the different cameras in random order. So you need a way to figure out which one is which. We have a parameter called "nickname" which you can set to a specific value on each camera, in your initscript.cfg, eg:

setpar nickname front

in the initscript.cfg of the microSD card that goes in your front camera.

Then your pi would first run a "getpar nickname" on each serial port to figure out which port is for the front camera vs rear camera, etc. Or you could also have this in your initscript.cfg:

setmapping2 .........(mode you need)

setpar serout USB

serout HELLO front

streamon

and on your Pi just look for messages that start with "HELLO". The serout command just sends its args to the serial output port.

Answered: Is there a way to install python libraries into jevois and run a piece of python code

Tue, 02 Jul 2019 23:24:19 +0000

Have you seen this one? No python there though:

http://jevois.org/tutorials/ProgrammerDlib.html

Are you trying to use this: https://pypi.org/project/face_recognition/

We will look into it as a good example of how to port a python module that has C++ dependencies.

Answered: I want to detect and recognition Waste (plastic, glass, paper) with JeVois camera with DNN

Tue, 02 Jul 2019 23:20:21 +0000

to detect (localize) and then recognize is more complex. You can try a custom version of YOLO.

I heard that AlexeyAB has the best tutorials on how to train a custom tiny-YOLO:

https://github.com/AlexeyAB/darknet#how-to-train-tiny-yolo-to-detect-your-custom-objects

https://github.com/pjreddie/darknet/issues/597

Answered: The problem with camera JeVois

Tue, 02 Jul 2019 23:15:47 +0000

Usually, you would only see this under very low light conditions, as the exposure is maxed out and the analog gain gets to the max.

In JeVois inventor, can you first select a module with 30fps, for example, TensorFlowEasy, and then go to the "Camera" tab and try manual exposure and manual gain, then switch back to auto exposure and gain (note that both parameters are coupled). Sometimes some module sets the camera to manual mode (e.g., the modules that do color blob detection), and it will stay like that until you set it back to auto (or restart the camera).

Answered: I need to create custom DNN for waste, garbage detection and classification - for SEPARATION

Tue, 02 Jul 2019 23:12:30 +0000

You can use this:

http://jevois.org/tutorials/UserTensorFlowTraining.html

but make sure you refer to the original tutorial from google, as the instructions for "toco" change depending on which version of Tensorflow you are using to train.

Or you can look for generic OpenCV tutorials about DNN training. Once you have a trained network that runs with OpenCV, you can load it onto JeVois, see for example

http://jevois.org/moddoc/PyClassificationDNN/modinfo.html

Answered: To activate different Arduino pin based on object position in image.

Mon, 17 Jun 2019 18:57:09 +0000

Yes, have a look at this one:

http://jevois.org/tutorials/UserArduinoBlink.html

Then, if you want different GPIOs to be activated, you just need a pinMode() statement for each pin in your setup() function, and later a digitalWrite() statement in the loop() function.

Answered: sending/reading a variable over serial

Wed, 29 May 2019 20:41:14 +0000

How about this one to start:

http://jevois.org/tutorials/ProgrammerInvSerial.html

http://jevois.org/tutorials/UserParseSerial.html

basically you use jevois.sendSerial() to send from JeVois to Pi. From Pi to JeVois you would implement parseSerial() in your JeVois module, see

http://jevois.org/doc/ModulePythonTutorial.html

Answered: Cross-compiling existing applications and libraries

Wed, 29 May 2019 20:35:43 +0000

Yes, we have many examples in the CMakeLists.txt of jevoisbase:

https://github.com/jevois/jevoisbase/blob/master/CMakeLists.txt

Usually the easiest is to figure out which .cpp files need to be compiled. Then just add them as source files in your JeVois module's CmakeLists.txt - that will get you both native and cross compilation for free.

Also see here for step-by-step instructions:

http://jevois.org/tutorials/ProgrammerDlib.html

Invoke and stream output video from a module using python code

Thu, 16 May 2019 07:31:12 +0000

Is there a way that a particular module can be invoked using python by sending messages to the camera?? If yes, is there a way to see the live video streaming output of the module using python without using any video viewing software. Also is there a way to get a snap shot to from darknet yolo module when a person is detected by the network??