Motion Target Tracking Based On Hue-Blob-Model

Motion Target Tracking is a crucial subject in Computer Vision which has important practical value in Intelligent Visual Surveillance, Military Affairs, etc. But the unstable features of the object result from occlusion are great challenge to algorithm's robust. We define a Hue-Blob-Model to solve that problem. And based on Hue-Blob-Model, we study and give two algorithms to solve rigid and nonrigid objects' tracking problems under occlusion.Hue-Blob-Model is restricted by Hue Probability Information and Hue-Blob's Location Information. It's a Mix Feature Model of region feature and point feature. There are multi-features for select in Hue-Blob-Model object tracking, which give great robust to algorithm.Aims to rigid object tracking problems under occlusion, a algorithm based on Hue-Blob-Model and Kalman filter is proposed. The algorithm uses dispersed independent Hue-Blob to hue-correlation-matching, and uses matching results to vote object's motion information. Kalman filter estimates Hue-Blob's mass center, and use it to minimize search region. Under occlusion use the reminder unoccluded Hue-Blob to match and vote. Experimental results indicate that it gives a robust tracking effect to rigid object under occlusion.MBMW-Shift algorithm is proposed to expand object from rigid to nonrigid. It based on non-parameter-estimate theory, develop Mean Shift algorithm to Multi-Hue and Max-Weight shift mode. An occlusion function is proposed based on relativity in Hue-Blobs' Motion. Experimental results indicate that it gives a better tracking effect to nonrigid target under occlusion.