opendatacam/opendatacam
An open source tool to quantify the world
What it solves
OpenDataCam provides a way to quantify the physical world by automatically counting and tracking moving objects in video feeds. It eliminates the need for manual counting in scenarios like traffic studies, allowing users to gather data on object movement and volume without constant human supervision.
How it works
The system uses machine learning and computer vision to detect objects within a camera feed or pre-recorded video. Once detected, it tracks the trajectories of these objects as they move across the scene. Users can define specific "counters" (such as lines or polygons) via a UI or API; whenever an object crosses these boundaries, the system records a count.
Who it’s for
It is designed for anyone needing to quantify moving objects, particularly those conducting traffic studies (e.g., modal-split or turn-counts), as well as users who want to deploy AI vision on edge devices like NVIDIA Jetson or in the cloud.
Highlights
- Broad Detection: Detects 50+ common object classes out of the box.
- Flexible Deployment: Runs on small field devices (NVIDIA Jetson), desktops, or cloud servers.
- Customizable: Supports training custom models and using custom neural network weights.
- Versatile Inputs: Works with real-time camera feeds or pre-recorded video files.
- Developer Friendly: Includes an API for extending functionality and exporting counter and tracker data.