基于智能手机的城市道路车辆即时识别

Real-time Recognition of Vehicle on Urban Road Based on Smart Phone

为及时了解路段交通状况,使用计算机视觉方法对城市道路上行驶中的车辆进行即时检测并分类计数,以便更便捷地对某路段交通状况进行定性定量的评估。使用普通的智能手机对城市道路拍摄视频,采用一种以YOLOv3为基础的目标检测和跟踪方法,实现了对车辆的分类识别、跟踪与计数。在配置好的Python环境下,YOLOv3算法可以快速准确地进行车辆分类检测,随后利用Deep Sort跟踪算法对各类车辆进行分类计数,得到了每帧画面中的各类车辆的数量。统计通过该路段的1000多辆车的型号,结合已有的汽车质量数据,对不同种类汽车重量的统计结果分别进行分布拟合,得到了每种车辆的车身质量的代表值。将质量代表值代入到识别出的计数结果中,对这一交通要道任一时段的汽车荷载进行了分析。结果表明:得出的汽车质量代表值和汽车数量分布,可以估算出每帧路面的车辆均布荷载,得到各个高峰期车辆均布荷载的分布区间;将其与规范规定的均布荷载标准值进行对比,发现至少在95%的情况下车辆均布荷载不会超过规定的均布荷载标准值,大部分时间符合规范要求;但在交通拥堵的情况下,车辆均布荷载也会超过规定的均布荷载标准值。因此,使用智能手机随时随地采集视频可快速准确地对车辆进行分类识别,并统计该路段的负载情况,对交通流量限制措施具有指导作用。

In order to understand the traffic situation of road section in time,the vehicles moving on the urban road are instantly detected,classified and counted by using the computer vision method,so that the traffic situation of a road section can be assessed qualitatively and quantitatively more conveniently.The videos on urban roads are shot by using ordinary smart phone.The classification,recognition,tracking and counting of vehicles are realized by using a method of object detection and tracking based on YOLOv3.In the configured Python environment,YOLOv3 Algorithm can quickly and accurately classify and detect vehicles,and then it use Deep Sort tracking algorithm to classify and count all kinds of vehicles to obtain the number of each kind of vehicles in each frame.The types of more than 1000 vehicles passing through the road section are counted.Combining with the existing vehicle mass data,the statistical weights of different types of vehicles are distributed and fitted respectively,and the representative value of the body mass of each type of vehicle is obtained.The vehicle load on the main road in any period is analyzed after substituting the mass representative value into the recognized counting result.The result shows that(1)The obtained vehicle mass representative value and the vehicle number distribution can be used to estimate the vehicle uniformly distributed load per frame on the road,and the distribution interval of the vehicle uniformly distributed load during each peak period can be obtained.(2)Comparing these values with the standard values of uniformly distributed load specified in the specification,it is found that at least 95%of the cases of uniformly distributed load will not exceed the standard value of uniform load specified in the specification,and they comply with the specification requirements in most of the time.(3)In the case of traffic congestion,the uniformly distributed load of vehicles will also exceed the standard value of specified uniformly distributed load.Therefore,using smart phones to capture videos anytime and anywhere can quickly and accurately classify and identify vehicles,and count the load on the road section,which can guide the measure of traffic volume limitation.