Efficient Color-Ingredient Particle Filter for Video Object Tracking

This paper proposes a new object model and a similarity measure method for particle filter. Based on cluster color histogram concept and similarity measure method, we analyze color ingredient and measure similarity using Euclidean distance, such that our approach can decrease memory consumption and increase processing speed effectively. Furthermore, in order to increase processing speed, we select the candidate particles based on the previous object segmentation. This can reduce the particle amount and speed up tracking operation. Comparing with the existing approaches, the experiments demonstrate that our method has batter performance even when moving objects go across complex scene. The proposed method can run comfortably in real time with 58 frames per second, and 4428 bytes memory consumption in average.