Improvements to Frank-Wolfe optimization for multi-detector multi-object tracking

This paper proposes a novel formulation for the multiobject tracking-by-detection paradigm for two (or more) input detectors. Using full-body and heads detections, the fusion helps to recover heavily occluded persons and to reduce false positives. The assignment of the two input features to a person and the extraction of the trajectories is commonly solved from one binary quadratic program (BQP). Due to the computational complexity of the NP-hard QP, we approximate the solution using the Frank-Wolfe algorithm. We propose several improvements to this solver affecting better minimization and shorter computations, compared to off-the-shelf BQP-solvers and the standard FrankWolfe algorithm. Evaluation on pedestrian tracking is provided for multiple scenarios, showing improved tracking quality over single input feature trackers and standard QPsolvers. Finally we present the performance of our tracker on the challenging MOT16 benchmark, being comparable to state-of-the-art trackers.