Multi-radar Detection And Tracking Technology For Intelligent Vehicles

Intelligent vehicle has become a hot topic for research around the world.Intelligent vehicle has been made up with a huge knowledge system and mainly need to solve four problems: positioning,perception,decision-making,and control.Intelligent vehicle can not only relieve traffic pressure,but also in the mining areas where working conditions are dangerous,driverless cars can replace people to perform operations,which increases the safety of workers.Lidar plays a crucial role in the perception of Intelligent vehicle,and has the characteristics of long distance detection,high precision,and low impaction of environment.The Lidar can provide enough road environment information for the Intelligent vehicle and can meet the real-time requirements of the Intelligent vehicle.It has important theoretical research value and engineering value.The main research contents of this article are as follows:(1)Analyze according to the installation position of the laser radar and its data characteristics,and propose radar data preprocessing based on multi-radar fusion.First,the multi-laser radar is used for network integration in the front area of the smart vehicle.Secondly,space-time alignment is carried out for different radar frequencies and positional relationships.Finally,a feature-based convergent voting method is proposed to detect obstacles and improve the detection rate of obstacles.(2)According to the characteristics of radar scanning fusion data,an information extraction algorithm based on multi-laser radar travelable region is proposed.Firstly,according to the characteristics of the radar return data combined with the data interval density distribution,we obtain the roadside point set and use the weighted Euclidean distance based K-nearest neighbor improved OPTICS algorithm to cluster the resulting roadside points.Then,use the least square method to fit the two sides of the road.Finally,the radar clustering in the travelable area is clustered by the improved OPTICS algorithm to detect the obstacle information required for the decision.Solve the problem that the obstacle information is inaccurate due to noise.(3)According to the multi-radar fusion data,under the condition that the inverse sensor model establishes a grid map,an algorithm is proposed to limit the Bayesian posterior probability by using fuzzy logic to correct the weight variable.Apply improved Bayesian inference to update grid status and use collision variables to detect dynamic obstacles.Finally,obstacle information and travelable area information are extracted by image processing means such as expansion,erosion,and improved connected area labeling.This method solves the problem that the original Bayesian inference method shows a strong lag when the grid state changes when the grid probability goes to the extreme.(4)In obstruction tracking,elliptical and positional dynamic tracking gates are established based on the characteristics of the vehicle’s target,eliminating interference echoes to reduce the number of joint events;secondly,the filtered echoes are clustered according to the target area;and finally,According to the targets in different categories and the multiple echoes in their dual thresholds,the respective types of association probabilities are calculated.The algorithm effectively reduces the number of joint events,and the goal of segmentation can simplify the number of split confirmation matrix to reduce the amount of calculation.