Data Association for an Adaptive Multi-target Particle Filter Tracking System

TL;DR

This paper introduces an adaptive multi-target particle filter tracking system that uses data association, state queues, and adaptive parameters to improve accuracy and occlusion handling in static scenes.

Contribution

The paper proposes a novel multi-target tracking approach combining data association, state queues, and adaptive parameters to enhance occlusion recovery and tracking accuracy.

Findings

Accurately tracks multiple objects in static scenes.

Handles partial occlusions effectively.

Recovers from occlusion events using adaptive parameters.

Abstract

This paper presents a novel approach to improve the accuracy of tracking multiple objects in a static scene using a particle filter system by introducing a data association step, a state queue for the collection of tracked objects and adaptive parameters to the system. The data association step makes use of the object detection phase and appearance model to determine if the approximated targets given by the particle filter step match the given set of detected objects. The remaining detected objects are used as information to instantiate new objects for tracking. State queues are also used for each tracked object to deal with occlusion events and occlusion recovery. Finally we present how the parameters adjust to occlusion events. The adaptive property of the system is also used for possible occlusion recovery. Results of the system are then compared to a ground truth data set for performance evaluation. Our system produced accurate results and was able to handle partially occluded objects as well as proper occlusion recovery from tracking multiple objects