Research On Intelligent Driving Path Planning Technology Based On Vector Map And Object Detection

Intelligent driving uses intelligent networking related technologies,not only providing safety assurance for people to travel,but also playing an important role in many engineering fields,receiving extensive research and attention.Path planning is one of the core links of intelligent driving,and the ability to conduct efficient and accurate path planning determines whether intelligent driving can fulfill specific needs and tasks.Mapping technology provides static environmental information for path planning,and perception technology provides dynamic environmental information for path planning.A mapping target detection planning path planning scheme has been formed,and its feasibility has been verified.In the aspect of intelligent driving map building,UAV map building and laser lidar map building have been tried respectively.Based on the principle of image matching and the aerial photography process,the unmanned aerial photography map was built for real scene modeling,and the resulting model was verified to obtain a point cloud map with an average error of centimeters.Using a unmanned vehicle equipped with a laser lidar to collect data,the NDT matching algorithm is used to match different frames of the point cloud,completing the construction of a laser lidar point cloud map.The obtained point cloud map is de-sampled,denoised,and segmented into ground point clouds.The voxel filtering algorithm is used for point cloud de-sampling.The effect of different voxel sizes on the proportion of filtered point clouds is studied.Statistical filtering algorithm was used for noise reduction,filtering out 5.4%of outliers.Using RANSAC algorithm to segment ground point clouds,the effect of different distance thresholds on the segmentation effect was studied.After point cloud map processing,a large number of invalid point clouds are simplified,and semantic information is annotated to generate a vector map.In the aspect of intelligent driving object detection,a lidar detection method based on improved euclidean distance clustering algorithm is studied.To address the issue of undersegmentation in European clustering algorithms,two methods were proposed to incorporate angle parameters into the European clustering judgment formula.Clustering algorithms that fuse distance and angle,as well as clustering algorithms that fuse distance and angle cosine values,were designed separately.The feasibility of the two improved algorithms was verified using partial point cloud models from the KITTI intelligent driving dataset and the self-made dataset,and the L-shape algorithm was used to generate a clustering boundary box as a visual display of the clustering results.The two verification methods have reached the same conclusion.The two validation methods obtained the same conclusion,and the two clustering algorithms can handle the undersegmentation problem of Euclidean clustering algorithms,and the clustering algorithm that integrates distance and angle cosine values performs better.In the path planning of intelligent driving,a path planning method based on improved ant colony algorithm is proposed.The Cauchy distribution function is introduced to set the adaptive pheromone volatilization coefficient,and the role of pheromone in different iteration stages is reasonably played;Genetic algorithm is introduced to ensure that the high-quality ants of each generation can be retained and improve the convergence speed;The idea of immune balance in immune algorithm is introduced to determine whether it is a similar ant by calculating the affinity between ants.The ant concentration is determined by the number of similar ants,and whether the ant updates pheromones is determined by the ant concentration and fitness performance,which improves the global search ability of ant colony algorithm.By solving the optimization problem of four benchmark functions,it is verified that the comprehensive performance of the improved ant colony algorithm is superior to other intelligent algorithms.The performance of the improved ant colony algorithm is compared with the self-made 20×20 and 30×30 grid maps,which proves that the improved ant colony algorithm has strong global search ability and fast convergence speed,and improves the efficiency and accuracy of path planning.The feasibility of the path planning scheme based on map building,object detection,and planning is verified by using Gazebo simulation software.