Meaning
Surveillance systems must process sensor data from multiple moving objects to maintain accurate trajectories over time. Multi-target tracking is the computational process of associating sequential sensor measurements with existing trajectories and estimating their future states. It operates as a core component in radar and lidar processing pipelines.
Data Association
Associating noisy sensor measurements with the correct object trajectories is a significant challenge when targets are closely spaced. Algorithms like the joint probabilistic data association filter calculate the probability of each measurement belonging to each active track. This calculation prevents the system from confusing adjacent objects or misallocating sensor returns.
Robust association is essential for maintaining track continuity in dense environments.
State Estimation
Predicting the future positions of objects requires mathematical modeling of their motion. Kalman filters and particle filters are used to update the estimated state based on new measurement data. This update reduces the uncertainty caused by sensor noise and temporary object maneuvers.
Track Maintenance
The system must handle the initialization of new tracks and the deletion of obsolete ones. When an object enters the sensor field, a temporary track is created. This track is confirmed after several consistent detections or deleted if no further measurements are received.
