基于分布场理论的行人检测算法研究

The Reasearch of Pedestrian Detection Algorithm Based on Distribution Field Theory

研究一种基于分布场理论的行人检测算法,首先获取深度图像;其次混合高斯背景建模并初步提取运动行人前景;然后从前景中提取分布场后计算最小矩形包络;最后提取行人。结合Open CV现有的GPU库对程序进行CUDA并行加速,并分别在PC平台和嵌入式平台NVIDIA Jetson TK1上实现。实验结果表明:经过CUDA并行化后,程序在PC平台和NVIDIA Jetson TK1平台上都获得了显著加速效果。

This paper introduces a pedestrian detection algorithm based on distribution field theory. The algorithm firstly gains depth image from depth camera,extracts foreground and calculate its distribution field,finally extract the contour and calculate the minimum rectangle envelope to achieve pedestrian detection results. According to the high parallelism of the algorithm, the program is accelerated by CUDA as well as the GPU library of Open CV and implemented on PC as well as NVIDIA Jetson TK1 platform. The experiment shows that this method is able to detect pedestrian from video sequence well, and the program speeds up significantly after CUDA acceleration on both PC and NVIDIA Jetson TK1 platform.