Research On Detection Of Targets Affecting Driving Comfort Based On Deep Learning

The emergence of autonomous driving technology may bring great convenience to people’s travel in the future.Its main method is to control the behavior of vehicles through the perception of driving environment,with high-precision maps and planning and decision-making algorithms,and to deliver passengers to their destinations safely.In recent years,the research and development of this autonomous driving technology has been in full swing at home and abroad,and good progress has been made.When the autonomous driving technology reaches the basic requirement of delivering passengers to their destinations safely,the requirement of driving comfort naturally arises.However,autonomous driving cars may encounter various road conditions that affect driving comfort on the way to their destinations,such as speed bumps,stones,bricks,puddles,manhole covers,etc.If such targets are not detected in time and appropriate deceleration or avoidance is made,it will bring unnecessary discomfort to passengers.In this paper,we analyze the obstacles in the road that may affect the driving comfort and make a relevant dataset,use a deep learning-based object detection algorithm to detect such targets,and design a mobile program for the detection of targets affecting the driving comfort to verify the performance of the detection algorithm in the road,the main research work of this paper is as follows:(1)We construct the dataset affecting driving comfort used in this paper,use Labelimg for annotation,and use data augmentation methods to expand the dataset for the problem of large differences in the number of targets instances that exist in the dataset.(2)To deal with the difficulties of the YOLOv5 model in detecting targets affecting driving comfort,Focal-EIo U is used instead of CIo U as the bounding box loss function to alleviate the problem of positive and negative sample imbalance in the training process while better locating the prediction box;the normalized Wasserstein distance is introduced into the loss function to improve the detection accuracy of small targets;the CA attention is added to the backbone network to improve the ability to extra the salient features of such targets;The WD-NMS algorithm is used to improve the Io U-based NMS algorithm to enhance the filtering effect of candidate box in the post-processing stage.(3)A mobile program for detecting targets affecting driver comfort is designed and implemented,and the improved algorithm proposed in this paper is deployed on the mobile side for verifying the performance of the improved model in the road.The experimental results show that the proposed improved YOLOv5 model increased from 83.4% to 85% compared with the original model m AP@0.5,m AP@0.5:0.95 increased from 43.9% to 45.9%,and also has a higher accuracy compared to the mainstream object detection algorithm.It indicates that the improved model proposed in this paper has a better detection effect on targets affecting driving comfort and ensures the real-time detection while improving the detection accuracy.