| Driving assistant system plays an important role in intelligent transportation.Therefore,the effective perception and processing of target dynamic information on road traffic has become a hot topic for researchers,and the research of vision system is particularly important.With the rapid development of deep learning,target segmentation algorithm can segment the target effectively.Combined with deep learning neural network,this paper studies the problem of target classification and distance detection for binocular vision images.The specific contents of this paper are as follows:(1)In the field of traditional image target detection,R-CNN algorithm has some limitations,such as poor classification effect and lack of detection ability.Aiming at the limitations of R-CNN algorithm,this paper uses mask R-CNN with the best performance as the detection and segmentation model.On this basis,a point to gravity weighted ranging method around the center of mass is designed,and the data set is constructed for target detection experiments.The experimental results show that the design method of human and environmental detection vehicle in this paper can achieve better results.(2)The features extracted by the traditional Mask R-CNN feature pyramid structure have the problem of insufficient representation ability.Aiming at the limitations of the feature pyramid structure,this paper designs a module including compression and excitation methods.Firstly,the module groups the features according to the channel,then compresses the excitation,then reweights and then merges,so as to increase the effective features output by the whole feature extraction part,and the directivity of these features to the target is better.Finally,the location information of the prediction frame is introduced,and the Non-Maximum Suppression algorithm of the location information is improved.Then the clustering method makes the model location information more accurate,and the confidence of the prediction frame is higher,so as to make the distance detection effect better.In addition,the improved model is tested on the data set in this paper.The experimental results show that the detection performance of FL R-CNN model is improved by 2% compared with Mask R-CNN model,and the ranging and segmentation are also better,which proves the effectiveness of the improved method in this paper. |
