Compact Data Association in Multiple Object Tracking: Pedestrian Tracking on Mobile Vehicle as Case Study

This paper presents a compact and fast data association which can be used in tracking-by-detection based multiple pedestrian tracking approaches. The goal is to make the data association simple and robust so that it can be really useful in a compact computing environment. This is realized by replacing the computationally heavy Bayesian based filter with the compact Median Flow tracker. Two layers of data association are proposed for fulfilling different requirements of the tracking tasks. Afterwards, we assess the performance of the algorithm by evaluating it on a standard dataset. Experimental results show that it improves processing speed upon state-of-art methods tremendously with only trading off limited performance.