The Research And Development Of Object Tracking System Based On Arm Embedded Platform

With the continuous development of social modernization, city population gradually increased, the problem of public safety, traffic safety, security and other issues have become the focus of public attention. The research of intelligent monitoring system in China is still staying in the primary stage, the theoreticaresearch of target-detection and target-tracking algorithm is backward,hardware platform infrastructure is not perfect, equipment operation efficiency is low, he performance is not stable.The paper integrates the ARM embedded development technology and computer vision technology, and also puts forward the target-tracking embedded system. Do useful attempt for embedded target-followup research, make it to be miniaturizated, real-timed.The main research work in this paper is:(1) The total design of the system hardware and software platformFirstly, the paper analyzes the core functional requirements of the system, completes the framework design of the system. Considering the using environment, development cycle, development costs and other factors of the system,we choose the mini2440 development board as the hardware platform of the system, the Linux operating system as the software layer.Then we set up a ARM-Linux development environment, and transplate the cutted Linux operating system to the development board; in order to deal with the video image information in the development board, complete cross compile and transplantation of Open Cv image processing library at the same time.(2) Improving SIFT target feature extraction algorithmThe SIFT feature has the characteristic of high dimension, large storage space occupied, the big complexity of the algorithm and poor real-time, we put forward the improved algorithm of SIFT feature. And we introduce the knowledge of the general process and the compressed sensing theory, that of the appearance target tracking model algorithm based on discrimination, to pave the way for the follow-up improved tracking algorithm.(3) Improved compressed sensing target tracking algorithmAccording to the disadvantages of the limited memory resource of embedded hardware platform, choose a compressed sensing target tracking algorithm, which has a smaller amount of calculation, good real-time target, as the research object. The algorithm would easily produce the drift phenomenon, when target is blocking or the appearance model changed greatly in case, so the robustness of algorithm is poor; in order to improve the algorithm, we put forward the adaptive learning rate adjustment algorithm, and another improved SIFT feature matching algorithm of target size adaptive correction, based on Bhattacharyya distance.(4) Comparative analysis of the experimentWe make the comparison of the improved algorithm and the original algorithm through the analysis experiment, the experiment shows that the improved algorithm is still able to track the target, when target is in gear, or the target appearance model changes in the objective circumstances. Compared with the original algorithm, the improved algorithm improve the performance significantly, in the accuracy and the robustness. Finally, write the improved algorithm into the ARM development board, and take the real tracking experiment.