Research On Vehicle And Pedestrian Detection Technology Based On Deep Learning

In recent years,the deep learning has been developed very quickly.The target detection algorithm based on deep learning is making breakthroughs,which has been widely used in Intelligent Transportation and other fields.The vehicle and pedestrian detection have been mainly studied in the Intelligent Transportation system.In actual application scenarios,vehicle and pedestrian detections are often affected by factors such as the size of the target object,the occlusion of the target and the environment,which results in lower accuracy.Therefore,in order to improve the detection effect,the research of a vehicle and pedestrian detection algorithm is very important.Aiming at the vehicle and pedestrian detection based on deep learning,corresponding solutions have been processed.The main studies are as follows:(1)A feature fusion model based on clustering is proposed.We optimizes the network structure of the SSD algorithm,by using the CBAM module as the attention mechanism modules and adding the CBAM module to the detection layers,which improves the accuracy of object detections and reduces the cost and the number of parameters.In addition,a feature fusion module is added to fuse features maps with rich semantic information and feature fusion maps with rich details so as to improve the information expression ability of the feature maps and improve the ability to detect small scale objects.(2)The k-means++ clustering algorithm is used to set anchor’s parameters of our model.At the same time Intersection over Union is used as the metric to avoid the influence of the candidate frame,which improves the training speed of the our model and the accuracy of positioning.It is proved by experiments that the anchor parameter based on the clustering is effective.(3)Aiming at the problem of low accuracy,the problems of poor detection of occluded target and unbalanced categories in the dataset have been addressed.The loss function and non-maximum the suppression algorithm is used to improve the accuracy by increasing the proportion of the difficult samples in the loss function so that the model improves the ability of detecting difficult samples.After that,the soft-NMS algorithm is used to process overlapping objects,which reduces the probability of the candidate frame around the target,thereby improving the algorithm’s ability to detect occluded objects and the accuracy.The experimental results show that our object detection algorithm improves the accuracy vehicle and pedestrian detection effectively.