Object Detect
Simple object detection using keypoint matching.
By Laurent Ittiitti@usc.eduhttp://jevois.orgGPL v3
 Language:   C++            Supports mappings with USB output:   Yes            Supports mappings with NO USB output:   No
 Video Mapping:   YUYV 320 252 30.0 YUYV 320 240 30.0 JeVois ObjectDetect

Module Documentation

This module finds objects by matching keypoint descriptors between the current image and a set of training images. Here we use SURF keypoints and descriptors as provided by OpenCV.

The algorithm consists of 4 phases:

  • detect keypoint locations, typically corners or other distinctive texture elements or markings;
  • compute keypoint descriptors, which are summary representations of the image neighborhood around each keypoint;
  • match descriptors from current image to descriptors previously extracted from training images;
  • if enough matches are found between the current image and a given training image, and they are of good enough quality, compute the homography (geometric transformation) between keypoint locations in that training image and locations of the matching keypoints in the current image. If it is well conditioned (i.e., a 3D viewpoint change could well explain how the keypoints moved between the training and current images), declare that a match was found, and draw a green rectangle around the detected object.

The algorithm comes by default with one training image, for the Priority Mail logo of the U.S. Postal Service. Search for "USPS priority mail" on the web and point JeVois to a picture of the logo on your screen to recognize it. See the screenshot of this module for an example of how it looks.

Offline training

Simply add images of the objects you want to detect in JEVOIS:/modules/JeVois/ObjectDetect/images/ on your JeVois microSD card. Those will be processed when the module starts. The names of recognized objects returned by this module are simply the file names of the pictures you have added in that directory. No additional training procedure is needed. Beware that the more images you add, the slower the algorithm will run, and the higher your chances of confusions among several of your objects.

With JeVois v1.1 or later, you do not need to eject the microSD from JeVois, and you can instead add images live by exporting the microSD inside JeVois using the usbsd command. See MicroSD card organization and files (last section) for details. When you are done addingnew images or deleting unwanted ones, properly eject the virtual USB flash drive, and JeVois will restart and load the new training data.

Live training

With JeVois v1.2 and later you can train this algorithm live.

First, enable display of a training window using:

setpar showwin true

You should now see a grey rectangle. You can adjust the window size and aspect ratio using the win parameter. By default, the algorithm will train new objects that occupy half width and height of the camera image.

Point your JeVois camera to a clean view of an object you want to learn (if possible, with a blank, featureless background, as this algorithm does not attempt to segment objects and would otherwise also learn features of the backgrouns as part of the object). Make sure the objects fits inside the grey rectangle and fills as much of it as possible. You should adjust the distance between th eobject and the camera, and the grey rectangle, to roughly match the distance at which you want to detect that object in the future. Then issue the command:

save somename

over a serial connection to JeVois, where somename is the name you want to give to this object. This will grab the current camera image, crop it using the grey rectangle, and save the crop as a new training image somename.png for immediate use. The algorithm will immediately re-train on all objects, including the new one. You should see the object being detected shortly after you send your save command. Note that we save the image as grayscale since this algorithm does not use color anyway.

You can see the list of current images by using command:

list

Finally, you can delete an image using command:

del somename

where somename is the object name without extension, and a .png extension will be added. The image will immediately be deleted and that object will not be recognized anymore.

Serial Messages

This module can send standardized serial messages as described in Standardized serial messages formatting. One message is issued on every video frame for the best detected object (highest score).

  • Serial message type: 2D
  • id: filename of the recognized object
  • x, y: standardized 2D coordinates of the object center
  • w, h, or vertices: Standardized bounding box around the object
  • extra: none (empty string)

Programmer notes

This algorithm is quite slow. So, here, we alternate between computing keypoints and descriptors on one frame (or more, depending on how slow that gets), and doing the matching on the next frame. This module also provides an example of letting some computation happen even after we exit the process() function. Here, we keep detecting keypoints and computing descriptors even outside process(). The itsKPfut future is our handle to that thread, and we also use it to alternate between detection and matching on alternating frames.

Custom module commands
list - show current list of training images
save somename - grab current frame and save as new training image somename.png
del somename - delete training image somename.png
ParameterTypeDescriptionDefaultValid Values
(ObjectDetect) winfloatpairWidth and height (in percent of image size, with valid percentages between 10.0 and 100.0) of the window used to interactively save objectsfloatpair(50.0F, 50.0F)-
(ObjectDetect) showwinboolShow the interactive image capture window when truefalse-
(ObjectMatcher) hessiandoubleHessian threshold800.0-
(ObjectMatcher) traindirstd::stringDirectory where training images areimages-
(ObjectMatcher) goodptsjevois::Range<size_t>Number range of good matches consideredjevois::Range<size_t>(15, 100)-
(ObjectMatcher) distthreshdoubleMaximum distance for a match to be considered good0.2-
Detailed docs:ObjectDetect
Copyright:Copyright (C) 2016 by Laurent Itti, iLab and the University of Southern California
License:GPL v3
Distribution:Unrestricted
Restrictions:None
Support URL:http://jevois.org/doc
Other URL:http://iLab.usc.edu
Address:University of Southern California, HNB-07A, 3641 Watt Way, Los Angeles, CA 90089-2520, USA