Adaptive Search Window for Object Tracking in the Crowds using Undecimated Wavelet Packet Features
Mohammad Khansari, Hamid Reza Rabiee, M. Asadi, Parisa Khadem Hamedani, M. Ghanbari · 2006
In this paper, we propose an adaptive object tracking algorithm in crowded scenes. The amplitudes of of undecimated wavelet packet tree coefficients for some selected pixels at the object border are used to create a feature vector (FV) corresponding to that pixel. The algorithm uses these FVs to track the pixels of small square blocks located at the vicinity of the object boundary. The search window is adapted through the use of texture information of the scene by finding the direction and speed of the object motion. Experimental results show a good object tracking performance in crowds that include object translation, rotation, scaling and partial occlusion.