Target Recognition Of Intelligent Vehicles Based On Binocular Vision Research On Distance Measurement System

In recent years,with the continuous improvement of technological level,cars that used to be luxury goods have entered thousands of households.The number of cars owned by households in China has increased significantly,causing traffic congestion and frequent accidents.Intelligent vehicles can better solve such traffic problems,and target recognition and ranging is the key technology in intelligent vehicles,so this thesis takes target recognition and ranging as the main research content,as follows:Firstly,for the problems of large scale changes and long time spent on analyzing targets in smart vehicle target recognition scenarios,an improved YOLOv5 s target detection and recognition algorithm is proposed in this thesis.Based on the YOLOv5 s network,in order to enhance the saliency of the detected targets,the channel and spatial attention modules are introduced into it to suppress the interference of irrelevant backgrounds;a bidirectional weighted feature pyramid structure is introduced into the feature fusion network,so that the network model can improve the feature fusion effect without increasing the computational cost;considering that the C3 Ghost module can lighten the network model,the original model CSP2＿X module in the neck network is replaced by the C3 Ghost module,which improves the computational speed while ensuring accuracy;the SIo U loss function is introduced to improve the convergence speed of the network model.In this thesis,KITTI public dataset is used to analyze the performance of the proposed algorithm,and the experimental results verify that the accuracy and speed of the proposed network model target recognition are better than traditional algorithms and meet the requirements of practical applications.Then,an improved Census transform stereo matching algorithm is proposed to address the problem that the accuracy of the parallax map obtained by the stereo matching algorithm in intelligent vehicle binocular ranging is easily affected by noise points.In the cost calculation stage,the mean value of the neighborhood of the central pixel point and the average amplitude of the neighborhood of the central pixel point are obtained,and the noise tolerance four-state information is introduced for matching cost calculation;the semi-global stereo matching SGM algorithm is used to complete the cost aggregation quickly and effectively;the winner-take-all strategy is used to select the better parallax,and then the final parallax map is obtained by parallax optimization and filling.In this thesis,the performance of the proposed algorithm is analyzed on the Middlebury testbed,and the experimental results verify that the accuracy and noise immunity of the proposed algorithm to obtain parallax maps are better than traditional algorithms.Finally,according to the method proposed in this thesis the target identification and ranging system is developed and designed by PyQt5,which can call the binocular camera to achieve real-time observation and image acquisition of the current scene and output the identification and ranging results.It is more intuitive and convenient for users to use.