Research Of The Object Tracking Method Based On Modified ASM

Visual object tracking gets great attention in the field of computer vision, and is widely applied to intelligent surveillance, human-computer interaction, and video coding, et al. The key technologies in tracking system are extracting and matching of the object features, both of which should be robust enough so that the following processing can implement effectively.Some traditional visual tracking methods are able to locate objects roughly, but can not extract object features accurately when the objects are non-rigid and deforming all the time. The methods based on deformable models focus on interpreting the contour of objects, obtaining more precise results when tracking non-rigid objects. The Active Shape Model (ASM) is a representative deformable model based method adapting to some shape variation while maintaining the shape specificity. It can extract object contours accurately in noisy and clutterd environment and with good robustness. This dissertation discusses the theories and applications of ASM on contour extracion in still image, and proposes a modified ASM based method for object tracking. The main contributions of the dissertation include the following aspects:First, amend the gray level model of ASM. As traditional ASM modelling part of the backgroud information, the model not able to adapt to the variation when object moves and background changes. To deal with this problem, the paper modifies the gray level model, improves the mathching algorithm by removing the background information when building gray level model and searching for the best match location by combining object inner gray information with strong edge feature. These improvements enhance the performance of contour extraction in speed and accuracy .Secondly, Propose an online extracting and updating mechanism. When tracking moving objects, it is hard to get the object gray level information beforehand, limiting the application of ASM to moving object tracking as traditional method use transcendental gray level model. Therefore online extracting and updating mechanism are important to reduce the dependence on transcendental condition.Thirdly, use Kalman filter to predict the object location. It improves the positioning accuracy of the initial model of ASM and the speed and accuracy of contour extraction in image sequences.Fourthly, design a tracking flowchat based on modified ASM method and Kalman filter. Verify the validity and scale adaptation of the proposed method by tracking the human body contour in this flowchat. And the experimental results show that the method proposed has better performance than traditional ASM in running time and matching accuracy.