面向道路交通场景的车辆违规占道实时检测

Real-time detection of vehicle illegal occupation in traffic scenarios

为了提高道路管理的智能化,面向交通场景搭建了车辆违规占道自动检测系统,对违规占道的车辆进行检测并自动识别其车牌。首先,以Yolov3目标检测器为基础,针对交通场景下的车辆检测,对检测器的主干网络进行改进,使得Yolov3更加适应现实的应用场景;然后,在检测到违规车辆后,基于卷积神经网络,对其进行车牌识别;最后,为了应对实际场景中硬件条件的限制,针对轻量的嵌入式GPU计算平台,在部署阶段对算法进行轻量化处理,来满足应用场景对算法的实时性要求。在真实交通场景数据中的应用结果表明:该方法能在满足实时性的同时,具有较高的准确性。

In order to improve the intelligence of road management in traffic scenarios,an automatic detection system was built to detect and automatically identify the license plate of vehicles for illegal encroachments.Based on the Yolov3 target detector,the backbone network of the detector was improved for vehicle detection in traffic scenarios,making Yolov3 more adaptable to realistic application scenarios.After detecting the offending vehicle,the license plate recognition was performed based on the convolutional neural network.Meanwhile,in order to cope with the limitations of hardware conditions in real scenarios,the algorithm was lightly processed in the deployment stage for the lightweight embedded GPU computing platform to meet the real-time requirements of the algorithm in application scenarios.The results of the application in real traffic scenarios show that the method not only meets the real-time requirements,but also has high accuracy.