Detection And Ranging Methods For Moving Targets In Complex Traffic Scenes Research And Applications

As China’s automobile industry continues to develop,the number of vehicles on the road continues to increase,and the resulting traffic safety issues have also increased year by year.Rear-end collisions account for a large proportion of these incidents,mainly due to drivers’ delayed reactions in unexpected situations,or when their vision is obstructed by poor environmental conditions.Traditional traffic management methods are difficult to effectively solve such problems.In recent years,with the continuous advancement of artificial intelligence technology,autonomous driving and intelligent transportation systems have become a major trend for future development.If efficient and fast detection of pedestrians,vehicles,and other targets in complex traffic scenes can be achieved using AI-based methods,and accurate distance estimation of these targets can be sent in real-time to the driver or automated driving system,it would be of great significance in reducing traffic accidents and improving traffic safety.Based on this,this article conducts research on the limitations of current detection and distance measurement methods for moving targets in complex traffic scenes.The research content is as follows:(1)In order to address the adverse external interference information such as rain,snow,fog,haze,and insufficient lighting that may occur in complex traffic scenes,a method for improving the AOD-Net model’s image enhancement is proposed.Firstly,the AOD-Net network structure is incorporated into a multi-scale network structure to improve the model’s defogging ability.Secondly,considering the insufficient lighting in practical application scenarios,the AOD-Net model is connected to a low-light enhancement module,and the decomposition network part of the Retinex-Net network in the low-light enhancement module is improved.Finally,the improved AOD-Net model with the multi-scale network structure is fused with the improved low-light enhancement model based on Retinex-Net,and experimental verification shows that the improved AOD-Net model has good image enhancement effects.(2)To address the low detection accuracy and high time cost of motion target detection in complex traffic scenes,a method for improving the YOLOv5 s network model’s motion target detection algorithm is proposed.Firstly,the YOLOv5 s backbone feature extraction network is replaced with the Shuffle Net V2 network,which has a lower parameter count,effectively reducing the model’s time cost.Secondly,to solve the problem of feature information loss caused by the replacement of the backbone feature extraction network,the Criss Cross Attention attention mechanism is introduced to effectively enhance the feature information of the motion target.Thirdly,the Upsample module in the Neck part of the YOLOv5 s model is replaced with the lightweight and general-purpose up-sampling operator CARAFE,which provides a larger receptive field for the network,thus improving the efficiency of object detection.Finally,the loss function of YOLOv5 s is changed to the Alpha Io U loss function,which solves the problem of differences between predicted values and true values during model training.The experiment shows that the improved YOLOv5 s has strong motion target detection ability.(3)To address the problem of low accuracy in measuring the distance of moving targets in complex traffic scenes,an improved monocular vision distance measurement model is proposed.For pedestrian targets,the model constructs a pedestrian distance measurement model that combines the camera tilt angle and the pedestrian’s position in front of the vehicle.For vehicle targets,the model constructs a vehicle distance measurement model that combines the camera tilt angle,the target vehicle’s position in front of the vehicle,and the camera’s blind spot.This reduces the impact of camera tilt angle,side targets,and camera blind spots on distance measurement accuracy,effectively improving the distance measurement accuracy of forward moving targets.(4)Based on the research results presented in this article,a motion target detection and distance measurement system is designed and implemented using the Py Qt5 framework.The entire system integrates three basic functions: image enhancement,target detection,and target distance measurement.It achieves the detection and distance measurement of moving targets in complex traffic scenes and verifies the effectiveness of the algorithms presented in this article.