Vehicle Identification And Stopping Detection In Highway Tunnel Based On Video Image

Real-time and accurate detection of parking incidents in highway tunnels is of great significance for reducing traffic accidents,reducing casualties and reducing economic losses.The space in the highway tunnel is narrow and closed,the traffic volume is large,and the speed is fast.When the vehicle suddenly stops,it is easy to cause traffic accidents.Especially when the vehicle carrying flammable and explosive materials is driving in the tunnel,the consequences will be more serious.The timely and effective parking detection can not only provide more rescue time for the injured people caused by the accident,but also notify other drivers on the road section to increase vigilance and reduce the occurrence of secondary accidents.In the current parking inspection work arrangement,the method of manually observing the monitor is still the main method.This method not only consumes a lot of human resources,but also has low management efficiency,which can no longer meet the needs of intelligence and efficiency in today’s traffic supervision.Therefore,this paper studies vehicle recognition and stopping detection algorithms based on video images for highway tunnels.Through research and analysis of different moving target detection algorithms,a stopping detection method based on mixed Gaussian background modeling is designed.However,this method has a large number of false detections.Therefore,a convolutional neural network(CNN)classification model is fused based on the detection method of moving objects to improve the accuracy of stopping detection.In addition,in order to further improve the accuracy,the nature of target detection is researched,instead of using traditional moving target detection,and instead based on deep learning target detection algorithm for vehicle recognition,and then complete stopping detection.In the highway tunnel scenario,the algorithm designed in this paper is tested.The test video is a segment intercepted from the video history storage file of the highway tunnel monitoring server,and the stopping event is included in the segment.In the experiment,comparing whether the moving target detection is fused with a convolutional neural network classification model,the results show that the positive detection rate of the fusion method is 84%,the false detection rate is 5%,and the missing detection rate is 11%;In the case of the network classification model,the positive detection rate is only 21%,the false detection rate is as high as 79%,and the missed detection rate is 11%.The positive detection rate of the fusion method increased by 63%,and the false detection rate decreased by 74%.In vehicle recognition based on deep learning,YOLOv3 has been improved from the aspects of data set production,clustering of prior frames,and optimization of network structure.Compared to the original YOLOv3’s 84.5% mAP on car,bus,and train in the VOC dataset,the newly designed model obtains 86.5% mAP,an increase of 2%.In addition,98.19%mAP was obtained on the self-made tunnel vehicle data set.On this basis,combined with the Deep SORT tracking algorithm,stopping detection was achieved.The positive detection rate was 95%,the false detection rate was 5%,and the missed detection rate was 0%.