Research On Target Detection Algorithm In Intelligent Transportation Application

With the improvement of people’s living standards,the number of cars is also increasing day by day.The Intelligent Transportation System alleviates the traffic pressure by coordinating the relationship among people,vehicles and roads.Among them,the detection of vehicle targets is one of the key points.However,in real life,complex weather conditions such as fog and rain often occur,resulting in the reduction of detection accuracy,Therefore,target detection in bad weather environment has become the focus of researchers.The target detection algorithm of YOLO series defines the target detection problem as a regression problem for training,positioning and classification learning at the same time,which can realize real-time detection.It is one of the widely used target detection methods based on the field of deep learning.In this dissertation,small targets are detected in foggy weather,and the detection accuracy is improved through algorithm improvement.The main work and innovations are as follows:（1）In this dissertation,the defogging image is comprehensively evaluated through time statistics and subjective evaluation.Finally,the non-parametric structure definition algorithm and the dark channel prior defogging algorithm based on guided filtering are used to restore the image,so as to obtain the real defogging data set,which has a certain effect on the improvement of subsequent target detection accuracy.（2）In the YOLO series target detection,YOLOv4-tiny is selected as the target detection algorithm studied in this dissertation,and an improvement strategy is proposed.Firstly,K-means++ clustering algorithm is used to cluster the frame center,and the appropriate cluster frame is selected to make the model more convergent;Then,Soft-NMS is used to screen the candidate boxes,which will improve the accuracy in detecting small target aggregates;Finally,three different attention mechanisms are introduced to improve the algorithm,strengthen the weight of useful features,and improve the detection accuracy under the condition of ensuring fast detection speed.In order to verify the effectiveness and robustness of the defogging algorithm and the improved target detection algorithm,the RESIDE RTTS data set is used for training and testing,and the VOC data set is used as the comparison data set to verify the effectiveness and stability of the algorithm.The experimental results show that the defogging data set has a certain improvement in accuracy compared with the original data set in different algorithms;On the defogging data set,the improved algorithm is more accurate than the original algorithm.