摘要
为了提高在复杂环境下检测遗留物体的准确度和实时性,提出了一种基于改进YOLOv2网络的遗留物检测算法。该算法在YOLOv2网络结构基础上结合残差网络,将浅层和深层特征多次融合,在基本不增加原有模型计算量和时间的情况下,提高了监控画面中检测小体积遗留物体的性能;同时以YOLOv2目标检测为基础,排除驻留行人和动物等非物体目标的干扰,并对目标筛选得到的可疑目标跟踪计时,停留时间超过阈值的目标标记为遗留物。以PETS2006和i-LIDS作为数据集进行实验,结果表明:该算法在提高遗留物检测准确度的同时缩短了处理时间,对人流密集的复杂环境抗干扰能力强。
To improve the accuracy and real time of detecting abandoned objects in complex environments,an abandoned object detection algorithm based on improved of YOLOv2 network was proposed.Based on YOLOv2 network structure,the algorithm combined the residual network and integrated shallow-layer and deep-layer features.In the case of no increase of the calculation amount and time of the original model,the performance of small-volume abandoned object detection in the monitoring screen was improved.Meanwhile,based on YOLOv2 target detection,the interference of non-objects such as pedestrians and animals was excluded,and the suspicious target gained by target screening was tracked and timed.The target with the time of stay exceeding the threshold target was marked as an abandoned object.PETS2006 and i-LIDS were used as data set for experiments.The results show that the algorithm improves the accuracy of the residue detection and reduces processing time,with strong anti-interference ability in the complex environment.
作者
张瑞林
张俊为
桂江生
高春波
包晓安
ZHANG Ruilin;ZHANG Junwei;GUI Jiangsheng;GAO Chunbo;BAO Xiaoan(School of Information Science and Technology,Zhejiang Sci-Tech University,Hangzhou 310018,China)
出处
《浙江理工大学学报（自然科学版）》
2018年第3期325-332,共8页
Journal of Zhejiang Sci-Tech University(Natural Sciences)
基金
国家自然科学基金项目(61502430
61379036
61562015)
浙江理工大学521人才培养计划
