Study Of Target Detection And Tracking System Based On Infrared Image Sequence

Infrared target detection and tracking system have characteristics of high precision, non-contact and anti-electronic jamming, was wildly used in military field such as optical reconnaissance and missile guidance. However, due to the complexity of the natural environment and the actual needs of engineering applications, further study of the infrared target detection and tracking system, in order to get a better detection and tracking effect, has been a challenging task, and also the majority of researchers continue to efforts.The algorithms for infrared dim target detection and tracking were studied in detail in this paper. In view of the problem of traditional fixed tracking window tracking method is easy to bring the interference, such as cloud edge information into the tracking window, which impacted the robustness of target tracking, a kind of adaptive tracking window algorithm was presented. In order to reducing the impact from the cloud edge information in tracking process and the amount of information needed to process, this algorithm dynamically changed the size of the track window by kalman filter while target tracking on the premise of detected the target. By being tested on the real infrared tracking platform, the proposed algorithm show it can still accurately tracking the target while clouds surrounding it, and the average processing time is about27ms/frame, approximately8 percent less than the average processing time of the traditional algorithm.The NCC (normalization cross correlation) measure is more stable than SAD (sum of absolute differences) measure when the illumination changes. However, it needs large calculated amount, which is its disadvantage. Therefore, a fast template matching algorithm of the NCC measure combing adaptive multilevel successive partitioning with the initial threshold estimation is proposed in this paper. The template image is progressively partitioned according to gradient values of different modules in the template image. Through the partition sequence, the summation of cross-correlation is divided into different levels, thus each bound of which will be obtained. Then, the eliminating approach of adaptive multilevel successive partitioning will be formed by getting the relation among different upper bounds with the application of the Cauchy-Schwarz inequality. Meanwhile, to accelerate the matching speed, the initial threshold, estimation is used to generate a larger boundary threshold, resulting in the elimination of lots of unmatched points in the initial search as well as the reduction of the number of searching points. The experimental findings demonstrate that the algorithm proposed in this paper has good robustness in the overall illumination changes, minor radiation interference and mild noise pollution, and the algorithm execution speed is superior to the traditional algorithmIn view of the problem of matching position error while target posture major changed or natural environmental changed, a template image update algorithm was proposed in this paper. It updated the template image while target posture or the external environment certainly changed, to obtain the new template representative of the target gesture. Experiments show that the template image update algorithm proposed in this paper can get the latest stance of the template image of the target, so it can ensure the correctness of the match tracking results compared to the matching algorithm that never updated the template image and the algorithm that updated the template image per-frame.Based on consideration of the real-time requirements of infrared target detection and tracking system, and the size of image data and complexity of the algorithm, the hardware based on DSP+FPGA architecture framework was proposed. The system mainly consists of image acquisition section, image processing section and system control components. Use DSP as the core devices for achieving the target detection and tracking algorithms, supported by the FPGA as a coprocessor together constitute the real-time image processing system. Based on it, the overall process of the system software and the performance index of the system were analyzed in detail.