Detection Of Illegal Driving In Bus Lane Based On SSD Model

With the coming of big data era and the improvement of computer performance,artificial intelligence has developed rapidly and has been extended to various fields.In recent years,deep learning has become a hot spot of artificial intelligence.It is also widely used in the field of computer vision.Target detection problems(known as target recognition or object detection)also have qualitative changes due to deep learning.Traditional target detection methods include three main steps:generating target suggestion box,extracting features based on the suggestion box,classification and border regression.However,the traditional target detection methods are very expensive in both computation and human resources.Therefore,it cannot realize real-time detection.Methods based on deep learning solve the above problems to a great extent.However,most of deep learning methods only focus on recognizing the basic instance object,lacking of target recognition capabilities with implicit logical relationships.Aiming at the practical problem of automatic detection about illegal driving in bus lane,this paper proposes three objective detection methods based on SSD model,which combine the detected vehicle information with lane information.The main work of this paper can be summarized as follows:firstly,collecting vehicle data of traffic violation information and designing a sorting software for manual sorting pictures.The advantage of this software is that it can load some basic deep learning object detection algorithm models,which can greatly reduce the labor effort.Secondly,this paper proposes three improved model,which combing the object detection results from SSD network and the lane information features from convolutional neural network.One is Two Stream Networks based on SSD and CNN(TSNSC).Two is Lane INformation Embedded SSD model(LineSSD),in which the different scales feature layers of SSD network add the same-scale lane information.Three is Double Channel CNN Combined SSD with Lane information(DC~3SL),based on the SSD network and double channel CNN.Finally,this paper implements the three models and compares them with each other.The experimental results show that models proposed in this paper perform better than SSD network.Especially DC3SL has obviously improved in the F1 score and the AUC.By introducing lane information into the SSD target detection model,this paper proposes several detection models for illegal use of specialized lanes.To a certain extent,it realizes the recognition and extraction of hidden logical relationship information in image,and broadens the range of object recognition deep learning models.In addition,the method proposed in this paper has a strong practical application background and good application potential.