Recently, the Kernelized Correlation Filters tracker (KCF) achievedcompetitive performance and robustness in visual object tracking. On the otherhand, visual trackers are not typically used in multiple object tracking. Inthis paper, we investigate how a robust visual tracker like KCF can improvemultiple object tracking. Since KCF is a fast tracker, many can be used inparallel and still result in fast tracking. We build a multiple object trackingsystem based on KCF and background subtraction. Background subtraction isapplied to extract moving objects and get their scale and size in combinationwith KCF outputs, while KCF is used for data association and to handlefragmentation and occlusion problems. As a result, KCF and backgroundsubtraction help each other to take tracking decision at every frame. SometimesKCF outputs are the most trustworthy (e.g. during occlusion), while in someother case, it is the background subtraction outputs. To validate theeffectiveness of our system, the algorithm is demonstrated on four urban videorecordings from a standard dataset. Results show that our method is competitivewith state-of-the-art trackers even if we use a much simpler data associationstep.
展开▼