摘要
在智能视频监控领域中,运动目标检测已经成为主要研究课题之一,针对传统的方向梯度直方图(HOG)算法并行化程度低等问题,采用了基于嵌入式GPU的并行化改进的运动目标检测方法,通过大数据量样本图片和实时采集视频进行检测验证,在GPU并行化模式下得到的处理速度都比CPU模式下处理速度高3倍以上,从而验证了经并行化优化的HOG算法检测速度明显提高,使系统整体的运行效率得到显著提升。
In the field of intelligent video surveillance, moving object detection has become one of the main research topics.IN order to solve the problem that traditional histogram of oriented gradients （HOG） algorithm has low degree of parallelism, it use the embedded GPU parallelization to improve the method of the motion target detection. Through large amounts sample of pictures and live video for testing, it verified the resulting speed-up in GPU mode are three times more than in CPU mode. The experimental results show that the detection rate and the overall operating efficiency of the system significantly improves by the parallel and optimized HOG algorithm.
出处
《电子设计工程》
2016年第22期134-137,共4页
Electronic Design Engineering
基金
江苏省科技厅政策引导类计划(产学研合作)--前瞻性联合研究项目(BY2015065-05)
