Target Recognition Based On LiDAR And Visual Sensors In Complex Environment

Target detection and recognition,as an important part of environmental perception,provides important data support for subsequent target tracking.However,due to the complexity of the actual road light,the recognition effect of vision sensors is unstable.Therefore,this article proposes a target detection and recognition system for complex environments based on laser radar and vision sensors.Through experimental analysis,it effectively improves the accuracy of target recognition.The main work of this paper is as follows:(1)A multi-period fusion algorithm based on the credibility weighted average of D-S evidence theory is proposed.Pearson correlation coefficient is introduced to calculate the credibility of the evidence bodies,and the weighted average evidence body is constructed for fusion to solve the conflict of evidence bodies caused by unclear target characteristics or sensor errors in complex environments.At the same time,multi-period fusion is introduced to make full use of the information of historical periods to improve the accuracy of recognition.Simulation results show that the algorithm can effectively solve the conflict problem and improve the confidence of target recognition.(2)The target recognition algorithm based on laser radar point cloud feature level fusion is studied,and the target object is extracted by preprocessing the original point cloud data obtained by laser radar.It is proposed that four groups of evidence bodies are constructed according to the four groups of feature data sets of basic characteristics,average curvature characteristics,contour characteristics,and internal characteristics.The single feature SVM classifier is used to construct the basic probability distribution function of evidence bodies.The constructed basic probability distribution function is fused by the multi-cycle fusion algorithm of credibility weighted average to realize the target recognition based on point cloud data.The point cloud data in KITTI dataset verifies that the algorithm can accurately identify the target category with high confidence.(3)The Retina Net algorithm is studied to detect and identify the targets in the image,to obtain the recognition results of the image.On this basis,a target recognition strategy based on multi-sensor decision-making level fusion is proposed.After spatiotemporal registration of laser radar point cloud and visual image,the target recognition is carried out respectively.Finally,the recognition results of visual sensor and laser radar are verified by each other,and further fusion is carried out to improve the confidence and accuracy of target recognition.