Complex Context Of Expansion Of The Target Sub-pixel Classification Tracking Method

The extended target recognition and tracking is a difficult and significant task for imaging tracking system under cluttered background. It is mainly confronted with two challenges: one is the cluttered background suppression and target recognition, andä»–the other is high-precision tracking and its stability for the extended target. The correlation tracking is a method based on comparability measurement that usually used for the extended target. The correlation method tracks target by finding the closed region between the template image and the current frame acquired from real scene, which has the advantage of low complexity, high stability, high accuracy and easy implementation.For tracking the extended target, we deeply analysed several typical methods and presented a new approach by improving the key steps in this dissertation. The simulation experiments have been done under MATLAB 7.0 which shew that the proposed method obtains good results with high accuracy and high stability for tracking the extended target under complex environment, comparing with the conventional methods. The main research work and the achievement obtained include the following contents:1. Pyramid-based hierarchy for the image shrinking controlled by the ratio of median filter and mean filter, which can improve the computation efficiency.2. Modified the correlation distance for MPC criterion, which full considered the effect of the similar pixel pair for correlation tracking.3. Improved the template updating strategy by introducing tracking judgment, and introduced the evaluation mechanism for tracking performance.4. Sub-pixel location was done by interpolating the image within a small window to obtain high accuracy with the half pixels.5. Designed a new approach to segment moving extended object in cluttered background. Experimental results of several real scenes show that the proposed method was more robust, higher precision, higher stability and had real-time capability. Meanwhile, the method shew good performance of ant-noise, ant-blur and ant-distortion. The image segmentation method was also presented in this paper that shew good efficiency.