Research On Target Detection Algorithm In Assisted Driving Scene

With the improvement of scientific and technological level,assisted driving system has developed rapidly.The auxiliary driving system not only brings a more comfortable driving experience to the driver,but also greatly improves the safety of road traffic,can effectively prevent traffic accidents and ensure the life safety.In assisted driving system,target detection technology is an important means of road perception,which has attracted the attention of scholars and engineering researchers.This paper discusses the application of image processing,deep learning and other technologies in the field of target detection.According to the accuracy requirements,real-time requirements and lightweight requirements in the auxiliary driving scene,this paper proposes effective improvements for existing models to realize the accurate detection of traffic targets under the condition of limited computing power.The contents are as follows:1.According to the requirements of lane line detection in auxiliary driving system,combined with image preprocessing,edge detection,Hough transform and other methods,a real-time lane line detection process based on image processing technology is designed,and the detection results are superimposed in the road video.Experiments verify the feasibility of the detection process.The proposed lane line detection method has good detection effect,meets the requirements of real-time and accuracy,and has certain anti-interference ability.2.Build a lightweight deep learning target detection model.This paper analyzes the requirements of detection system in assisted driving scene,and designs improved routes for vehicle platform with limited computing ability.One route is to combine the YOLOv4 algorithm with the lightweight neural network Mobile Net,replace the CSPDark Net-53 feature extraction network with Mobile Net network,and replace the original ordinary convolution with deep separable convolution,which reduces the parameters and obtains great detection speed.Another improvement route is to add the attention mechanism module to the residual structure based on the YOLOv4-tiny algorithm,and use the Cm BN method to replace the BN method of the original CBL components,so as to improve the feature extraction ability.3.Build a traffic light target detection model based on improved YOLOv5 algorithm.Aiming at the demand of traffic light target detection in assisted driving,this paper improves the C3 module and feature extraction network of YOLOv5 model,replaces the feature fusion mode of C3 module with add operation from concat operation,eliminates the Focus module in the feature extraction network,optimizes the parameter configuration,avoids frequent slicing operation.The improved YOLOv5-lite model not only maintains the detection accuracy,but also improves the processing speed,which is easier to deploy on the vehicle platform.