The heart of OpenTrafficCam.
OTVision is a collection of algorithms to generate trajectories of road users from traffic videos. Most of the algorithms have been developed in previous open source projects. We modify them and link them with our own developments into a working pipeline.
The current pipeline consists of three core functionalities of OTVision: convert, detect and track.
graph LR subgraph OTVision["OTVision:"] direction LR conv(<b>convert</b>\n\n...raw video\nfiles to mp4) det(<b>detect</b>\n\n...road users\nin single\nframes) tr(<b>track</b>\n\n...detected road\nusers over\nmultiple frames) conv --> det --> tr end tr .-> traj[/.ottrk\nTrajectories/] vf[/.mp4\nVideo Files/] .-> det rvf[/.h264\nVideo Files/] .-> conv
- Can be used without programming knowledge
- Conversion of .h264 video files to other formats (using ffmpeg)
- Detection (joint localization and classification) of road users using state-of-the-art AI object detection models in single video frames (currently using YOLOv5 by Ultralytics)
- Tracking of detected road users over multiple frames (currently using the IOU-Tracker by Bochinski et al.) and over multiple videos.
- Result: Trajectories in pixel coordinates
Last update: February 24, 2023