基于多目标跟踪的交通异常事件检测  

Traffic anomaly detection based on multi-target tracking

作　　者：高新闻[1,2] 沈卓 许国耀 封玲 Gao Xinwen;Shen Zhuo;Xu Guoyao;Feng Ling(Institute of Mechanical&Electrical Engineering&Automation,Shanghai University,Shanghai 200444,China;SHU-SUCG Research Center for Building Industrialization,Shanghai University,Shanghai 200072,China;SHU-UTS SILC Business School,Shanghai University,Shanghai 201800,China;Shanghai Pujiang Bridge&Tunnel Operation Management Co.Ltd.,Shanghai 201315,China)

机构地区：[1]上海大学机电工程与自动化学院,上海200444 [2]上海大学上海大学—上海城建建筑产业化研究中心,上海200072 [3]上海大学悉尼工商学院,上海201800 [4]上海浦江桥隧运营管理有限公司,上海201315

出　　处：《计算机应用研究》2021年第6期1879-1883,共5页Application Research of Computers

基　　金：上海市科委项目(18DZ1201204)。

摘　　要：针对传统异常行为自动检测方法的准确率和稳定性无法满足多变视频检测需求的问题,将最新的目标检测网络YOLOv3与目标跟踪算法相结合,通过对基于SORT多目标跟踪框架的改进,对检测目标的级联匹配采用了融合运动与外观特征的指标,以适应实际高架桥梁道路监控的情况。然后利用改进的多目标跟踪算法,对城市高架道路监控视频中的目标进行跟踪,配合相应的轨迹判别规则实现对视频中出现的行人、停车和车辆变道的交通行为异常情况的自动判别,具有较高的判别精度,可以达到实际应用目的。Aiming at the problem that the accuracy and stability of the traditional automatic detection method of abnormal behavior cannot meet various demand of video detection,this paper combined the latest target detection network YOLOv3 with the target tracking algorithm.Through the improvement of the multi-target tracking framework based on SORT,the cascade matching of detection targets adopted the index of fusion of motion and appearance features to adapt to the actual situation of via-duct road monitoring.Then,it used the improved multi-target tracking algorithm to track the target in the monitoring video of urban elevated road.With the corresponding trajectory discrimination rules,it realized the automatic discrimination of abnormal traffic behaviors of pedestrians,parking and vehicles changing lanes in the video.It has high discrimination accuracy and can achieve practical application purposes.

关 键 词：交通异常检测 行为检测 多目标跟踪 YOLOv3算法 SORT框架 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]