Object tracking and trajectory recognition using improved CAMSHIFT and Hidden Markov Model

Vision-based motion object trajectory recognition is currently one of the hot spot of scientific research. This paper describes a method for extracting and classifying two-dimensional motion in an image sequence based on trajectory. In the trajectory feature extract stage, as the traditional CAMSHIFT algorithm can´t exclude the non-target objects which have the similar color space, we proposed an improved method. The image obtained from the background subtraction of GMM background model do AND operation with the binary image based on threshold segmentation of HSV color space, then, the result image (G-H image) do AND operation again with the color probability distribution image in CAMSHIFT algorithm. We can avoid the interference of non-target through this improved CAMSHIFT. In the next stage, threshold segmentation was used to get the start and end points of the trajectory. In the final stage, the trajectory is recognized by using Left-right Banded model, Baum-Welch algorithm and Viterbi algorithm. Experiment result shows this system is satisfied for applications.