The OpenPTrack team is excited to announce the release of OPT V2 (Gnocchi). The code can be found on the OPT V2 Github page, and provides new, fundamental features (GPU acceleration required):
- Object Tracking. For the first time, OpenPTrack will track objects in addition to the human body. V2 will add the capability to track objects with the machine learning-based YOLO, allowing integration of custom-training sets.
- Pose recognition. OPT will now add the capability to annotate person tracks with poses recognized from a pre-trained set, using machine learning-based skeletal tracking which utilizes the OpenPose library.
- Enhanced real-time person detection. OPT V2 includes a new experimental person detector using convolutional neural networks; this improves performance in detecting people sitting, lying down, heavily occluded, etc.
The Gnocchi release also updates the underlying software stack to Ubuntu Linux 16.04 LTS and Robot Operating System (ROS) Kinetic Kame. (Though some components will still work with Ubuntu 14.04 and ROS Indigo, we will no longer be able to test/support them.)
Along with the new version of OpenPTrack, we will be completing a major overhaul of the wiki that will support and detail OPT V2 features, and will be releasing a dockerized version of OPT to simplify deployment.
(Click images for larger views.)
Multiview Pose Tracking
Object Detection
Zed Camera, CNN-based People Detection