A Bayesian Filter for Multi-view 3D Multi-object Tracking with Occlusion Handling
This addresses multi-object tracking in crowded, occluded scenes for applications like surveillance, but it is incremental as it builds on existing Bayesian filtering with a new occlusion model.
The paper tackles online multi-camera multi-object tracking in 3D with occlusion handling, proposing an algorithm that achieves linear complexity in detections and scales with camera numbers, evaluated on WILDTRACKS and a new crowded dataset.
This paper proposes an online multi-camera multi-object tracker that only requires monocular detector training, independent of the multi-camera configurations, allowing seamless extension/deletion of cameras without retraining effort. The proposed algorithm has a linear complexity in the total number of detections across the cameras, and hence scales gracefully with the number of cameras. It operates in the 3D world frame, and provides 3D trajectory estimates of the objects. The key innovation is a high fidelity yet tractable 3D occlusion model, amenable to optimal Bayesian multi-view multi-object filtering, which seamlessly integrates, into a single Bayesian recursion, the sub-tasks of track management, state estimation, clutter rejection, and occlusion/misdetection handling. The proposed algorithm is evaluated on the latest WILDTRACKS dataset, and demonstrated to work in very crowded scenes on a new dataset.