Research On The Design Of Movingobject Detection System Based On Heterogeneous Multi-core Architecture

Moving object detection is an important branch of computer vision and is widely used in various fields of social production and living. Moving object detection is the initial stage of video analysis. The accuracy and instantaneity of the detection will directly impact on the subsequent stages. The improvement on the accuracy, robustness and instantaneity of moving object detection algorithm has important significance of reality.The optical flow algorithm is a widely used moving object detection algorithm which can accurately estimate the motion information and surface structure of the moving object when knowing nothing about the scene. The optical flow algorithm can be applied in both dynamic scene and static scene while it keeps high accuracy and robustness. The shortage of optical flow algorithm is that the algorithm is not real-time because of the high time complexity and large scale of calculation, which limits the application of the algorithm. Fortunately, the optical flow algorithm has inherent parallelism which can be used to implement the algorithm parallelly. The researchers proposed solutions based on heterogeneous architectures such as GPU, DSP and FPGA to performed the algorithm in parallel and pipeline in order to improve instantaneity of the optical flow algorithm.The main content of this paper is to perform and accelerate the process of moving object detection by using a heterogeneous multi-core architecture based on ARM-FPGA. The key point is to assign tasks for ARM and FPGA according to their respective functions. The ARM processor is assigning to get a video frame from the camera, to display the result and to analyze the results of image processing implemented by FPGA. The FPGA is assigning to perform the algorithm of moving object detection in parallel and pipeline. The ARM and FPGA work together to accelerate the implementation of moving object detection. This paper proposes a software-hardware collaborating platform based on the ARM-FPGA heterogeneous architecture deployed on the Zynq-7000 SOC introduced by the Xilinx.Inc. The ARM subsystem and the FPGA subsystem communicate with each other via the AXI-BUS and shared-memory inside the SOC. The communicating experiments showed that the internal communication is very fast and the speed can meet the real-time requirements. In this paper the Horn&Schunck optical flow algorithm performed the detection of moving object. The algorithm is implemented in parallel and pipeline based on the FPGA subsystem according to its inherent parallelism. The experimental results proved that the time consumption of software processing is 17 s while the software-hardware collaborating processing is 0.42 s when the input image sequences is 584x388 pixels. The time consumption of the moving object detection is effectively decreased so that the real-time of the system is significantly improved.