Study On Object Tracking In Video Images

Target tracking as an interdisciplinary cutting-edge technology, integrating image processing, pattern recognition, artificial intelligence, automatic control and other knowledge from different areas. It has broad application prospects in military guidance,navigation,safety monitoring,transportation,video coding,medical diagnostics,weather analysis,astronomical observations and so on. Researching on tracking algorithm has important practical significance and theoretical value.This paper comprehensively discusses various tracking methods and divides them to several classifications. We pay more attention to the most representative method—Mean Shift and Particle Filter. In this paper we elaborate in detail the steps,features and the development of each algorithm in recent years. According to Mean Shift algorithm, we enumerate the various optimization methods when the scale of the target transform,the light conditions change and present a automatic feature selection method using ratio-log graph. Experimental results show that the algorithm in many scenarios has a good robustness and improve the tracking accuracy, can better adapt to lighting changes, background interference, partly obscured or the quality of color is very poor. For the particle filter, we described in detail the theoretical basis and research against the defects raised by the improved algorithm.Aiming at the vehicle tracking applications, we analyze the characteristics of road control problems, and the achievement of all key procedure in the process. We select several tracking video for experiment. In the target recognizing section we use two methods—background subtraction and Gaussian Mixture Models. Compare the results to analyze the advantages and disadvantages of each method. Due to occlusion and noise and other practical problems, experimental results still have many deficiencies, require further study to resolve.