Object Tracking

Execution Priority

Prioritize calling this skill's scripts/run.py entrypoint first. Only write custom code after the script result is unsuitable or the script cannot cover the requested task.

Input Resolution

If detections are missing but the user provided frames or a video, do not ask the user to provide detections and do not write tracking code. Call upstream skills in order: frame-sampling/scripts/run.py when frames are missing, then object-detection/scripts/run.py, then this skill. Use the detection labels implied by the task, such as person for people, fight, crowd, or behavior analysis.

Atomic CLI

Run this skill directly with its own script. The script does not call other skill scripts and does not depend on shared src, tools, or registry modules.

python object-tracking/scripts/run.py --detections-json <detections.json> --camera-id <camera_id> --config <config.json> --output <tracks.json>

Parameters: --detections-json, --camera-id, --association-distance-pixels, --stationary-distance-pixels, --config, --output.

Workflow

1. Sort detections by timestamp.
2. Associate same-label detections across frames.
3. Build track trajectory, start/end times, duration, last bbox, and confidence.
4. Classify movement state as stationary or moving.

Available Implementation

This skill is implemented as an atomic standalone script in its own scripts/run.py. The script contains the executable logic for this skill and must not import shared src, tools, or registry modules.

Inputs

• camera_id
• detections
• Optional tracks to pass through precomputed tracks

Outputs

Returns tracks, each with track_id, label, start_time, end_time, duration_seconds, trajectory, movement_state, confidence, last_bbox, and evidence_frame_ids.

Failure Modes

The current implementation tolerates empty detections and returns an empty track list.

Constraints

Do not apply ROI rules or emit event candidates.