Research On Intelligent Vehicle Environment Sensing Technology Based On Camera And Lidar Information Fusion

With the development of automobile intelligence,self-driving vehicle appears in people’s vision in recent years.The development of self-driving vehicle technology can bring convenience to people’s life and reduce the occurrence of traffic accidents.As an important part of self-driving technology,intelligent vehicle environment sensing technology has undoubtedly become the focus of research at home and abroad.It is of great significance to study how to use sensors to obtain more accurate environmental information of road scene.Based on the above recognition,this paper studies the dynamic target detection and recognition in road scene based on camera and lidar,and provides sufficient environmental information for self-driving vehicles.First of all,based on the image data collected by the vision sensor,the road image is preprocessed by graying and Gaussian filtering,and different detection methods are used to detect the lane line.The detection of straight road is completed by using threshold masking method and Hough transform.For curved road,the detection of curved road is realized based on the detection method of sliding window polynomial fitting.Secondly,according to the deep learning algorithm,the dynamic target detection in the road scene is realized,including vehicles and pedestrians.The method of Haar feature+ Ada Boost classifier is used to train vehicle detector to realize vehicle recognition.The method of Hog feature + SVM classifier is used to train pedestrian detector to complete pedestrian recognition.The deep learning framework of Tensor Flow is used to build convolutional neural network to realize the detection of multiple dynamic objects such as pedestrian and vehicle in road scene.Then,based on the characteristics of laser 3D point cloud and Mean-shift clustering algorithm,the dynamic obstacles in road scene are clustered.By analyzing the detection characteristics of vision sensor and laser sensor,combined with the advantages of both,a dynamic target detection method based on monocular camera and lidar information fusion is researched.Using the software platform of Autoware,the two sensors are jointly calibrated,and the joint calibration parameters are obtained.The fusion of threedimensional point cloud to two-dimensional image is realized by using the calibration results,and then the dynamic target detection is realized by using the Yolo-v3 algorithm and Euclidean clustering respectively for the fused information.Finally,using the test platform of the unmanned electric wheelchair,online experiments are carried out on the detection algorithm and fusion method used in this paper in the campus road.Through the statistical results and comparative analysis,the accuracy of target detection based on information fusion is obviously better than that relying on a single sensor,which verifies the feasibility and effectiveness of the algorithm.