Research On Road Scene Recognition Algorithm Based On Deep Learning

The increasing use of automobiles in people’s daily lives has brought great convenience to people’s travel.However,at the same time,traffic safety issues have become increasingly prominent.The problem of drivers is a major factor that causes traffic accidents.Therefore,major scientific research institutions hope to assist human-driven driving through research and development of autopilot technology to reduce traffic accidents caused by driver factors.In actual road scenes,the accuracy and speed of object recognition is a major challenge for researchers due to the adverse effects of changes in weather illumination,object obstructions,and motion blur.This topic is based on deep learning algorithm and designs a road scene recognizer,which has good recognition effect for common objects such as pedestrians,vehicles,animals and traffic signs in road scenes.This paper first did a lot of data set preprocessing work.In addition to the simple manual video capture,in order to increase the number and diversity of data,the image was subjected to operations such as radiation transformation,blur processing,and brightness transformation.The number and diversity of the dataset images;then the ResNet-34 network model was adopted,and the network structure was improved and the network training speed was accelerated.The improved network was used to train the traffic sign classification model and the final model was classified.The correct rate reached 96.602%.Finally,the overall scenario of the road scene recognizer was designed using the SSD+VGG network and R-FCN+ResNet network.The SSD algorithm was used to identify large-size objects such as pedestrians and vehicles based on the VGG network.The use of R-FCN algorithm based on the ResNet network to identify small-scale traffic signs,the final good recognition of the effectiveness of the proposed algorithm.