基于HOG特征提取的车辆检测系统的FPGA实现OA

Developing vehicle detection controllers in embedded systems is a complex task due to limited hardware resources.The Histo-gram of orientation gradients has been widely used to detect and match features.However,the HOG algorithm is computationally deman-ding and its implementation in embedded systems requires more efficient methods.An optimised pipeline architecture for HOG feature ex-traction is designed and the system is deployed on an FPGA for vehicle detection.Firstly,a hardware acceleration module for HOG feature extraction is designed on the FPGA side of Zynq.Secondly,a support vector machine module is designed on the ARM side for feature clas-sification.Finally,the AXI4 bus is used to connect the FPGA and ARM for data interaction between hardware and software.Experimental results show that the system accelerates the recognition process while consuming less hardware resources than similar existing work.On the Xilinx Zynq-7020 platform,the system proposed is able to reach a processing speed of 64 FPS using an 800×480 pixel image resolution,while the classification accuracy is 91.86%,which is of good practical value in vehicle control applications.