Research And Application Of Region-based Convolutional Network In Target Detection And Shadow Environment

With the continuous development of the field of computer vision,various algorithms have become more and more capable of recognizing images.Shadows,as interference factors in images,will have an impact on feature extraction.A target detection model in a shadow environment is proposed,and the model is used to combine the common requirements of smart driving to build a usable street view detection system.The main work includes the following aspects:(1)Build a new shadow detection module based on the attention mechanism and refer to the residual neural network architecture using the well-known shadow data sets SBU and UCF,and use BER as the evaluation parameter to filter out the shadow semantics for the image.The shadow detection module based on the attention mechanism has a good effect,laying a foundation for the street scene target detection model in the entire shadow environment.(2)Using the Mask-RCNN model for target detection and semantic segmentation based on regional convolutional neural networks,using a custom Cityscapes street view data set for training,modifying the classification category,reducing training loss,and building a street view target detection and semantic segmentation module.Finally,the shadow detection module and the street view detection module are combined to perform shadow recognition and street view classification detection on street view pictures.The final detection result uses m AP and m AR as the evaluation criteria.(3)On the basis of research,a street view detection system in a shaded environment was designed and implemented.The trained detection model was loaded into the system,and pictures or videos were uploaded to complete the detection easily.Through the demand analysis of the street view detection system and the design of various functional modules,the entire street view detection system has been completed after development and testing.Compared with the existing target detection models in the shadow environment,the street view target detection model proposed in this paper focuses on the existence of shadows in the image,magnifies its visible impact,detects the specific location of the shadow,and avoids its impact on the target detection task.Various experimental results show that the model has improved in all indicators,which can effectively improve the recognition effect of target detection in a shadow environment.