Location:Home > Engineering science > Detection Technology and Automation > Research of Motion Object Detection and Tracking in Video Image Sequences

Research of Motion Object Detection and Tracking in Video Image Sequences

Downloads: []
Tutor: TangMingHao
School: Donghua University
Course: Detection Technology and Automation
Keywords: object detection and tracking,Kalman filter,Mean Shift,particle filter,artificia
CLC: TP391.41
Type: Master's thesis
Year:  2010
Facebook Google+ Email Gmail Evernote LinkedIn Twitter Addthis

not access Image Error Other errors

Motion object detection and tracking is the crucial subject of computer vision. Many fields need the technology¡¯s application, such as human-machine interface, medical diagnosis, intelligent video surveillance and intelligent robots, etc. Many subject¡¯s knowledge is needed in the research including image processing, pattern recognition and artificial intelligence.This paper mainly does the following research in basis of the work on basic theory and key technology of motion object detection and tracking.In the aspect of motion object detection, the paper proposes the algorithm of self-adaptive threshold which improve on the judgment threshold of object pixel and background pixel, according to the common detection method based on single Gaussian background model. The paper also proposes a method of parameter initialization that can save storage space and reduce the computational complexity, and the reduction of computation in Gaussian model. In the end of the part, a detection method and its improvement are proposed based on the fusion of background model and frame difference analysis. The simple frame difference is replaced by the correlation coefficient of the frame image, the background¡¯s update is on the basis of all the same gray value pixels¡¯ average variation to reduce the computation, and a simple and efficient method of eliminating light¡¯s change is provided.In the aspect of motion object tracking, according to the basic color histogram, the result better reflects the human visual characteristics after introduce HSV color model instead of common RGB color model. For the problem that basic Kalman filter¡¯s tracking algorithm can¡¯t solve nonlinear and non-Gaussian operation, the tracking algorithm based on extended Kalman filter is proposed to transform nonlinear problem into the uniformly accelerated motion.To improve the target model description of basic Mean Shift object tracking algorithm, not only the color histogram is weighted by kernel function, but also the edge histogram is introduced and the two feature information are fused. For the problem of tracking window scale invariant, the adaptive scale algorithm is proposed too.According to the particle degradation of object tracking algorithm based on particle filter, the paper introduces the thought of artificial immune, put the target¡¯s histogram as antigen, the particle set¡¯s histogram as antibody, the similarity coefficient between the target¡¯s and particle set¡¯s histogram as the affinity between antigen and antibody, by the way of colon to promote the large affinity particle and inhibit the small affinity particle, use the mutation means to set large variation for small affinity particle and set small variation for large affinity particle, so the system can converge to the global optimal solution rapidly.
Related Dissertations
Last updated
Sponsored Links
Home |About Us| Contact Us| Feedback| Privacy | copyright | Back to top