JeVois  1.9
JeVois Smart Embedded Machine Vision Toolkit
Share this page:
Changes and new features in JeVois 1.9

JeVois 1.9 brings the following new features:

  • Introduced the new JeVois Inventor graphical user interface which makes programming JeVois using full OpenCV and full Python 3.6 much easier than ever before.
  • Bumped to OpenCV 3.4.2, with all contrib modules including deep neural networks (DNN)
  • YOLO v3 is now supported by DarknetYOLO, faster, more accurate, less memory, and 80 object classes.
  • New modules DetectionDNN and PyDetectionDNN show how to use object detection + recognition deep neural networks using the OpenCV DNN framework, in C++ and Python. Supported network formats are Caffe, TensorFlow, Darknet, and Torch. Examples are provided which include:
    • OpenCV Face Detector, Caffe model
    • MobileNet + SSD trained on Pascal VOC (20 object classes), Caffe model
    • MobileNet + SSD trained on Coco (80 object classes), TensorFlow model
    • MobileNet v2 + SSD trained on Coco (80 object classes), TensorFlow model
    • Darknet Tiny YOLO v3 trained on Coco (80 object classes), Darknet model
    • Darknet Tiny YOLO v2 trained on Pascal VOC (20 object classes), Darknet model
  • Added pre-trained MobileNet v2 networks to TensorFlowEasy
  • New standardized messages for object recognition results, and for object box detection + recognition defined in Standardized serial messages formatting and used in all deep network modules.
  • Two new tutorials explain how to decode these messages on Arduino, with sample Arduino code: UserArduinoBlink and UserArduinoYOLO
  • Added support for different camera sensors. ov7725 is working great, still work in progress for ov2640. More on that coming in the next release.
  • Added a frame counter in the main loop, and access to it using jevois::Module::frameNum(). Modules that need to keep track of frame number do not need to implement their own counter anymore and can use this one.
  • Added parameter serstamp to jevois::StdModule to optionally prepend all standardized serial messages with frame number, time, frame+time, etc
  • Added parameter serlimit to jevois::Engine to allow limiting the number of serial messages sent per video frame, to avoid overloading a serial link.
  • Added parameter sermark to optionally mark start and end of each frame. Useful if one needs to know when a frame has been processed but no results were found and hence no other serial messages were sent.
  • Added standardized messages for object recognition and using them in the neural network modules (TensorFlow, Darknet)
  • new command-line interface command: shell (allows to run Unix commands on JeVois)
  • new command-line interface commands to get machine-oriented info about the camera: cmdinfo, caminfo, modcmdinfo, paraminfo
  • Fixed bug in RoadNavigation which was sending incorrect data when operating with no USB output.
  • Enabled swap support in the JeVois kernel. Swap is not turned on by default to not slow down boot time, but, if needed, one can easily create a swap file and activate it using the shell command of the JeVois command-line interface.