Design and Realization of Real-Time Motion Target Detection System Based on ZYNQ

Motion target detection algorithms have been maturely implemented on traditional PCs, but they are large in size and high in power consumption. In order to be used in spacerestricted scenarios, a motion target detection system based on ZYNQ hardware acceleration is designed. First of all, we customized the QDMA IP core for the problem that the AXI VDMA IP core provided by XILINX requires multi-parameter configuration, which is not flexible enough to be used. This IP core can be hooked to the bus to realize the read and write access to the memory address space, compared with the VDMA IP core, the custom IP core only needs to set 50% of its configuration items to read and write data. Secondly, the system is modularized by combining the characteristics of ZYNQ’s hardware and software synergy. The algorithm module including grayscale processing, median filter, image difference, morphological filter is processed by the programmable logic side which is good at parallel computing for pipeline operation, and the camera configuration, system memory read/write control is processed by the ARM side which is good at serial execution of complex programs, which not only improves the processing speed of the algorithm but also significantly shortens the development time compared with a single processing method. Experimental tests show that the system can effectively detect moving targets in real time within a certain distance, and compared with the traditional PC platform using the same processing algorithms, it can still run smoothly at a resolution of more than 30fps in $1280 \times720$ under the condition that the power consumption is only 1/22 of the CPU power consumption (2.039W).