Optical Flow-based Image Object Tracking Method

The sequence image tracking is an extremely important problem in the field of computer vision. The optical flow methods have been put into a better application in detection and tracking of moving objects for sequence image. But the methods have the following faults: heavy computational burden and weak anti-noise ability. Furthermore, it can not track the moving objects with high speed. The feature-optical-flow method has the better solutions for the faults.The thesis researches the feature-optical flow method on the tracking moving objects;the main works are as follows:1. The theoretical analysis and the experimental comparison are given for corner detection based on the curvature algorithm and the Plessy corner detection algorithm. For the two corner detection algorithms, three quantitative evaluations (stability, reliability and anti-noise) show that the performance of corner detection algorithm based on the curvature is better than the Plessy corner detection.2. The classical optical flow methods based on difference theory and based on five-point restraints are researched. The performance comparisons of the two algorithms on extraction of moving objects are given. The simulation results confirm that five-point restraints optical flow algorithm has the better precision with less computational burden.3. The five-point restraints optical flow algorithm is applied to tracking moving objects. Simulation results demonstrate that when tracking moving vehicles this algorithm can only obtain the outlines instead of the whole picture.4. The problems of moving object image tracking based on feature-optical-flow are researched. For solving problems when objects are tracked real-time and circumrotated, an optical flow clustering rule is produced, and the moment feature is introduced, a novel moving object tracking method based on the moment feature and the feature-optical-flow is proposed by utilizing the moment feature match method.5. The flow of the novel algorithm is described in detail. The image tracking simulation results indicate that the novel tracking algorithm has better performance on computational burden, tracking precision and anti-circumrotation ability. Specially, when the object rotates within 30 degree, the method can track the object steadily.