Research On Dynammic Obstacle Recognition And Obstacle Avoidance Method For Driverless Cars

Driverless cars are typical intelligent mobile robots,which involve many disciplines.They are important research directions in today's artificial intelligence and automotive fields.The obstacle avoidance system is the top priority for research on driverless cars.Obstacle avoidance system refers to the process of detection and recognition of obstacles to successfully avoiding obstacles in complex environment.The collision avoidance of moving obstacles has always been the difficulty of this research.In this thesis,the identification of moving obstacles,collision prediction and collision avoidance path planning are studied in depth and verified by experiments.The specific contents of the study are as follows:(1)Accurate and real-time recognition of obstacles is the basis for obstacle avoidance systems to perform collision avoidance,but current image recognition algorithms have limitations in real-time and accuracy.Aiming at the existing problems,a moving obstacle recognition algorithm based on convolutional neural network is proposed.The method integrates global features on the basis of layer by layer extraction of detail features,and uses global features to correct the limitations of detail features in the recognition of the target is too small,and finally uses Softmax to classify.Experiments show that the improved convolutional neural network has higher accuracy in identifying moving obstacles.(2)The prediction of the trajectory and collision point of moving obstacles is the precondition of collision avoidance system to avoid collisions.In this paper,a three-dimensional raster map is applied and a collision prediction method is proposed,which not only retains the characteristics of the traditional grid maps to accurately represent the occupation of obstacles,but also can express the relationship among time,driverless cars and moving obstacles.In the environmental modeling of three-dimensional grid maps,motion obstacles are classified into three types:linear type,curve type,and intersection type,and the predicted collision points are obtained according to the predicted results.(3)Planning a path that can avoid obstacles is the key to collision avoidance.In the thesis,a path planning algorithm based on the improved A*algorithm is proposed to solve the problem of poor real-time performance and sharp planning trajectory of the traditional A*algorithm.This algorithm uses exponential decay as the weight of the valuation function,and uses the method of reducing the weight rapidly with the distance between the target points shortened to improve the real-time performance.To improve smoothness,the algorithm increases the number of searchable neighborhood nodes from countable to infinity,and uses Bezier curves fitting the planned path.Using the real road data,the theoretical content of the study is applied to the simulation platform for experimental verification.The experimental results fully prove the effectiveness of the specific methods studied in this thesis.Finally,the advantages and disadvantages of the research content are summarized,and the future research is prospected.