融合Camshift与YOLOv4车辆检测算法  

Fusion of Camshift and YOLOv4 Vehicle Detection Algorithm

作　　者：胡习之[1] 魏征 周文超 HU Xizhi;WEI Zheng;ZHOU Wenchao(School of Mechanical and Automotive Engineering,South China University of Technology,Guangzhou Guangdong 510641,China)

机构地区：[1]华南理工大学机械与汽车工程学院,广东广州510641

出　　处：《机床与液压》2021年第11期70-74,共5页Machine Tool & Hydraulics

基　　金：国家自然科学基金项目(51975217)。

摘　　要：作为one-stage代表作的YOLO系列最新算法,YOLOv4在检测速度和精度相比于YOLOv3均有提升,但是YOLOv4在视频流的检测速度上仍有提升的空间。提出一种融合Camshift和YOLOv4的车辆目标检测算法。算法的流程为:首先计算图像的差异值哈希值,然后利用哈希值来判断当前帧图像与上一帧图像的相似度,当相似度小于阈值,则交给YOLOv4算法进行检测,并将检测结果传给Camshift作为其初始化跟踪窗口;当相似度大于阈值,则由Camshift算法来进行跟踪。最后在实际道路上采集的数据进行算法检测,检测结果表明融合算法的有效性。As the latest algorithm of YOLO series,which is the masterpiece of one-stage,YOLOv4 has improved the detection speed and accuracy compared with YOLOv3,but YOLOv4 still has room for improvement in the detection speed of video streams.A vehicle target detection algorithm integrating Camshift and YOLOv4 was proposed.The process of the algorithm was:the difference value hashing value of the image was calculated,then the hash value was used to determine the similarity between the current frame image and the previous frame image;when the similarity was less than the threshold,it was handed over to the YOLOv4 algorithm for detection,and the detection result was passed to Camshift as its initialized tracking window;when the similarity was greater than the threshold,it was tracked by the Camshift algorithm.Finally,the data collected on the actual road was tested by the algorithm.The test results show the effectiveness of the fusion algorithm.

关 键 词：YOLOv4算法 CAMSHIFT算法 差异值哈希算法 算法融合 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]