Study On Detection And Tracking Of Small Target In Infrared Imagery

Detection and tracking of moving infrared (IR) targets (especially small IR targets) is one of the key technologies in infrared reconnaissance warning. For the particularity of infrared imaging, the processing of IR image is different from visual image processing. IR images of distant small moving targets obtained from real weapon systems under complex backgrounds have extremely low signal to noise ratio, non-repeatability signature, limited detail information, which make detecting or tracking targets even harder.Based on the requirements of weaponry preliminary research topic-"Detection and Tracking of Infrared Targets on Military Car Reconnaissance System" and the property of IR target image, IR preprocessing, target detection and tracking algorithms are studied and the plan of detection and tracking target is provided. Comparing good denoising filter and targets edge detection method, Canny edge detection algorithm based on wavelet denoising is adopted and good results of the method are shown. For the disadvantage of Mean Shift algorithm towards occasions in IR target tracking, such as target occlusion, drastic change of image intensity, high ego-motion of the sensor, the relevant solutions are proposed. First, the center of target located by target detection is used to initialize Mean Shift algorithm automatically. Second, to overcome the problems related to the changes in the target feature, target distributions are computed by the cascaded kernels. The decision whether to change the target model is based on the distance measure computed from the likelihood of target and candidate distributions. In cases when the target is occluded or moves more than the operational limits of the tracking module, Kalman tracking and the target center locating detection are used to modify tracking iteration. Thus, automatic detection and tracking of IR target in real-time is achieved and the proposed method is more robust in target tracking.