Design And Implementation Of Object Tracking System Based On ARM-FPGA

Object tracking is one of the most challenging fields of the computer vision. It involves many high-level topics of computer vision such as optical flow, feature extraction, template matching, detecting and learning. The tracking problem is such a challenging task because it aims to solve the following questions: scene illumination changes, complex object shapes, non-rigid or articulated nature of objects, partial and full object occlusions, multi-target tracking.Because of the complexity of the algorithm, it requires high-performance computer to carry out the costly process which highly limits the application of the subject. As the development of the computing power for the CPU for some embedded platform. Some simple tracking tasks have been tested in some high-performance embedded platforms.Firstly, the thesis analyses the main embedded solutions for computer vision which include DSP, ARM and FPGA. Based on the analysis of the three solutions, balance the efficiency of developing, the performance of the CPU and the BOM costs of the solutions. This thesis adopt the FPGA-ARM based architecture as the hardware design. Design and accomplish a complete test set of hardware and software to run the object tracking task. The system includes the image signal collection board, algorithm testing platform and the software running on the host computer to facilitate viewing and controlling the embedded system.Subsequently, this thesis analyses the main theories that is acclaimed for their efficiency to solve the tracking tasks and pick the STC(Spatio-Temporal Context Learning) algorithm as the base theory. A implement of the algorithm is completed and tested on the notebook computer.Finally, the performance of STC is tested. While testing the algorithm on the platform, the short coming of the algorithm is analyzed and amended using GMM for stationary cameras.