Research On Methods Of Specific Bject Tracking

The research on motive object tracking technology is an active issue incomputer vision, motion image coding and many other fields, which has avery wide range of applications in both military and civilian aspects likeintelligent transportation, medical diagnosis, and machine navigation, i.e. Inthis paper, we studies some common object tracking algorithms, and thenfocuses on analyzing and improving the Camshift tracking algorithm and itsfurther improvement combined with Kalman filter for some specific rigidbodytargets which have certaincolor or edge information. After the detailedanalysis of the principle and implementation process of traditional Camshift,we propose corresponding improvement solutions for interferences fromillumination variation, similar color object or background, and possibleocclusion.1.The traditional Camshift algorithm builds a one dimensionalhistogram relying on hue from HSV color space, which may lead to thefailure of tracking when interfered by illumination variation and similarcolor object or background. To address this problem, we propose animproved algorithm based on a three dimensional histogram, which is builtusing hue and saturation components from HSV space and edge gradientfrom object’s shape information. As compared with background histogrammodel, these three components are weighted adaptively in the objecthistogram model to improve object tracking performance under backgroundinterference.2.The multi-feature blended Camshift algorithm based on athree-dimensional histogram then combines Kalman filter to predict theinitial position of the search window more accurately. This improved algorithm significantly benefits the tracking accuracy and the times ofiterations. In addition to the Kalman filter in occlusion conditions, ourimproved algorithm is amended accordingly to increase the noise immunityin tracking process.3.We also complete the calculation for pan and tilt angle control of thePTZ camera with its SDK development platform, shifting the target centroidto the center of the image feasibly. Experiment results demonstrate that thePTZ camera can be well controlled in tracking targets actively and timely.