Skip to main content

Object Tracking

The object tracking service in the Chariot platform provides real-time object tracking capabilities on top of model inference data. It allows you to maintain persistent object identities across successive inferences (e.g., across video frames or streaming data) by creating tracks for detected objects and updating their state over time. In particular, the object tracking service listens to model inference outputs (such as object detection results) and links detections that belong to the same moving object into a continuous track with a unique ID. This is crucial for many applications in which you want to know where each object moves frame-to-frame or any scenario requiring consistency of object identity over time and position rather than treating each inference independently.

By using the object tracking service, you can turn your model's raw detection events into higher-level insights: Who/what is this object and where is it going? The service abstracts the complexities of data association, state estimation, and track management into a configurable component of the Chariot platform. You can create tracker instances (each managing tracks for a sequence of inferences), feed them detection results, and query the Inference Store for tracked objects.

By following this documentation, you should be able to configure the object tracking service to suit your application, understand how it is maintaining object identities over time, and use the Python SDK to integrate tracking into your automated workflows. The object tracking service, together with Chariot's Inference Servers and Inference Store, enables pipelines such as real-time monitoring (e.g., continuously tracking vehicles in a live camera feed and raising alerts when certain conditions are met) or offline analytics (e.g., processing a collection of videos to gather trajectory data for objects of interest).