Research And Implementation Of Vehicle Detection Technology Based On Radar And Vision Fusio

The vehicle detection system is an important member of today’s smart cities and intelligent transportation systems,and is also a real-time,accurate,and efficient comprehensive intelligent management system.In vehicle detection systems,radar and camera are the main perception tools that can obtain information such as speed,distance,and angle of multiple perceived objects in the sensor illuminated area,and can directly collect image information of objects in the area.Therefore,both are widely used for information perception in various complex scenes.In intelligent transportation scenarios,sensors may encounter issues such as overlapping,obstructing,or low detection rates for some vehicles(large trucks)when detecting vehicles.At the same time,extreme weather such as night,rain,and snow can also increase the difficulty of sensor detection.At this time,a single sensor may not be able to successfully complete the perception task,and it is necessary to apply lightning vision fusion technology for vehicle detection.Therefore,this article mainly focuses on innovative calibration,registration,and fusion detection algorithms in the fusion process of radar and video sensors.The innovative algorithms have been simulated and tested on a large number of datasets and measured data.The specific work is summarized as follows:(1)A joint radar and video calibration algorithm based on a robust and fast Pn P method is proposed to address the issues of poor external calibration accuracy and high computational cost of radar and video sensors during the calibration process.By analyzing the shortcomings and shortcomings of existing Pn P methods,focusing on the selection of control points and mapping relationships during the mapping process,as well as the elimination of errors during the solving process,and considering the uncertainty of observation and propagation during the mapping process,the RF-Pn P algorithm was designed.This algorithm can combine radar information for control point selection,thereby solving the problem of large radar interest regions.It can synchronously iterate the optimal solution and some mapping errors during the process,further improving the accuracy of joint calibration.Finally,comparative simulations were conducted under three conditions: standard,quasi singular,and planar.The RF-Pn P algorithm showed a reduction of about 15% and 8% in average translation error and average rotation error compared to advanced algorithms in the same period.The experimental results showed that the RF-Pn P algorithm has a more robust and fast calibration effect.(2)Aiming at the registration distortion caused by the missing of outlier and overlap values in the point cloud data of mine view,an adaptive weight vector ICP registration algorithm based on surface reconstruction is proposed.Firstly,it analyzes the shortcomings of the existing ICP algorithm,such as the small convergence domain,and the inability to process the outlier and overlapping values in the point cloud data.Based on this,RAWV-ICP algorithm is proposed.The algorithm first considers the changes in normal vectors and curvature to enhance the algorithm’s constraints,and proposes a new mathematical model based on reconstructing local surfaces.This model has a wider convergence domain and better convergence effect compared to the original point-to-point constraint model.Secondly,for outlier and overlapping values,RAWV-ICP algorithm can allocate appropriate weights to optimize the noise and outlier interference of the whole system.Finally,this chapter conducted extensive experiments on the proposed algorithm on noisy,abnormal datasets,and real datasets.Compared with advanced algorithms in the same period,all performance indicators(RMSE,return value)ranked first in terms of overall performance score.At the same time,this article innovatively proposes two performance indicators to evaluate the ICP registration effect: the elimination value of overlapping regions,the alignment value of estimated interior points and ground truth interior points.Compared with the advanced algorithm RAWV-ICP in the same period,the proposed algorithm has improved by about 12% and 9%,respectively.Through experiments,it has been proven that the proposed algorithm has better performance compared to other advanced algorithms.(3)A target detection algorithm CT-EPn P based on lightning fusion is proposed to address the common issues of overlapping object misdetection,missed detection,and multiple detections in the joint detection process.This algorithm uses radar and camera data for sensor data fusion,and adds a center fusion algorithm on top of the EPn P algorithm.During mapping,the truncated cone method is used to compensate for the radar information on the associated image.CT-EPn P can fully combine the depth,rotation,and velocity attributes of the detected object.After simulation verification on VOC and Nu Scenes datasets and derivation of relevant mathematical formulas,it has been proven that the CT-EPn P algorithm improves the detection rate of overlapping objects by about6% and reduces the false detection rate by 8%.The experimental results show that CT-EPn P can have better detection performance.In summary,during my research,I encountered problems such as a large amount of lightning fusion data,low calibration accuracy,and poor handling of data anomalies.Based on the above issues,I innovatively proposed a lightning joint calibration algorithm,a lightning registration algorithm,and a lightning fusion detection algorithm,which successfully solved the errors caused by the accumulation of object shape changes and iterative redundancy during the lightning external parameter calibration process,as well as the missing point clouds during the lightning registration process The registration distortion caused by outlier and overlapping values,and the false detection and missing detection of overlapping objects in the process of lightning vision fusion detection.The proposed method has good application prospects in the field of intelligent transportation,and the technology in this article has been applied in Beijing Chuansu Microwave Co.,Ltd.with practical value.