Research On Video-based Road Abandoned Objects Detection Technology

Abandoned objects on the road are one of the potential risks affecting traffic safety.Timely and accurate detection of abandoned objects on the road is a very important link in traffic safety management.The present abandoned object detection algorithm has the problem of low precision.Facing the demand for abandoned object detection on the road,the paper researches the algorithm and detection scheme of abandoned object detection on the road based on the video.The main research contents of this paper are as follows:(1)Two kinds of abandoned object image data sets for abandoned object detection in highway and tunnel scenes were constructed.One is a total of 20723 samples of abandoned objects,including boxes,bags,tires,damaged and scattered objects,sticks,cloths,and other debris in various scenes.The data set can be applied to the training and testing of the classification network.The other includes two kinds of pavement image data sets based on top view,namely,top view data set based on background modeling and top view data set based on target detection network.The data set contains 64400 sample pictures,which can be applied to the training and testing of anomaly detection network.(2)The abandoned object detection method of pavement was designed based on image perspective conversion and the Dense Net network.For the design of abandoned objects on pavement,the Gaussian mixture background modeling algorithm based on image perspective conversion was used to extract suspected abandoned objects,and the impact of perspective projection on abandoned object images and abandoned object detection was weakened based on camera calibration.Based on the Dense Net network,the road abandoned object detection method was designed.The proposed method was used to detect suspected abandoned objects extracted by the top view background modeling algorithm,and the targets mistakenly extracted by the suspected abandoned object extraction algorithm were removed.The experimental results show that the proposed method can not only improve the detection accuracy of abandoned objects on pavement but also detect the real size of abandoned objects.(3)The anomaly detection network model for abandoned object detection on the pavement was designed.Based on the image input methods of two kinds of anomaly detection network models,namely the top view of pavement background modeling and the top view of target detection network,two kinds of anomaly detection network models of abandoned objects on the pavement were designed.The model is trained based on the anomaly detection network data set designed in this paper.By improving the output layer of the anomaly detection network,the abandoned object can be accurately selected on the image and the real size of the abandoned object can be detected.(4)In view of the demand for pavement abandoned object detection,the paper designs the pavement abandoned object detection scheme and its application in combination with the videobased pavement abandoned object detection algorithm.The design scheme includes video image acquisition,abandoned object detection,detection result processing,and abandoned object event alarm.In the scheme,an optimization algorithm of abandoned object detection results based on an object detection network is designed for parking and vehicle congestion scenarios which are prone to false alarms of abandoned objects.In order to prevent the same abandoned object from being reported repeatedly in a short time,the abandoned object event alarm strategy was designed.All kinds of real traffic videos are tested to verify the effectiveness of the scheme.The video-based pavement abandoned object detection algorithm studied in this paper can be used for highway abandoned object detection,and can better solve the problem of low abandoned object detection accuracy.