Opendatacam

OpenDataCam is an open source tool that uses computer vision to detect, track and count moving objects in video. It recognizes more than 50 common object classes out of the box, including cars, trucks, buses, bikes, motorcycles and pedestrians, and users can train their own model for anything else. It is especially popular for traffic studies such as modal split and turn counts. Users define counters through a web interface or the REST API, and each time an object crosses one the system records the time, type and direction. Objects that cross several counters can be followed, and trajectory analysis shows how pedestrians cross a square or how vehicles cross lanes. Input can come from USB cameras, IP cameras, RTSP streams and YouTube live streams, or from pre recorded video files. The setup is meant for city planners, researchers and developers. OpenDataCam is distributed as Docker images and runs on anything from an NVIDIA Jetson Nano in the field to a data center server, provided it has a GPU and CUDA. Privacy is a stated design goal: processing happens on location, only metadata such as paths and counts is saved, and no visual data is stored or leaves the device. Data can be downloaded through the interface or the API. The site publishes no prices. It points professional users and businesses to separate professional offers and services.