Design Of Unmanned Vehicle Forward Collision Warning System Based On Multi-sensor Fusion

Nowadays,with the development of society by leaps and bounds,the number of cars owned by the people of our country is constantly increasing.Unmanned driving technology is booming,and many driving assistance functions have been developed to reduce car accidents and other hazards caused by car problems.On a global scale,road safety issues are the key issues studied by many scholars,especially the problem of vehicle collisions.An efficient method is needed to solve them.On the basis of meeting the accuracy requirements of the software system,the real-time performance of the control system is continuously improved.This reduces the response time of the system and ultimately improves the driving safety of the vehicle.At present,the image-based target detection algorithm is very mature,but the detection algorithm based on sensor fusion data association needs to be improved.This paper will adopt the current mainstream image-based target category detection algorithm,propose a new type of laser point cloud ground point separation algorithm and subsequent improved clustering algorithm and compare it with traditional methods to improve the performance of the algorithm;at the same time combine A variety of sensor correlation algorithms fusion processing the perception results,and then analyze and verify the efficiency of the algorithm;the final text designs the unmanned vehicle system,and on the basis of the designed unmanned vehicle software and hardware system,this article The designed fusion algorithm and the anti-collision warning submodule were simulated and verified in real vehicles.The main content of this article is as follows:The first chapter introduces the research significance of the multi-sensor fusionoriented unmanned vehicle collision avoidance warning system and the current research status at home and abroad,puts forward the main content involved in this article,and briefly summarizes the specific technical routes involved in this article.The second chapter discusses the theories related to the image-based target category perception system and the laser point cloud-based target state perception system,and combines the KCF algorithm and neural network model to improve the inference speed;this paper proposes A laser point cloud ground point separation algorithm based on vertical lines,and compared with two traditional algorithms.These two methods are the ground point separation algorithm based on random sampling consistency and the ground point based on ground plane fitting.Separation algorithm.The effectiveness and efficiency of the algorithm proposed in this paper are verified by comparison;the non-ground points are clustered by improving the traditional European clustering algorithm,and then the center point of the target and its state are estimated.The third chapter performs feature-level data fusion on the perception results of Chapter 2,respectively aligns and synchronizes time and space;derives the internal parameters of the sensor coordinate system and its rigid body transformation,and then integrates the point cloud,image,and vehicle coordinates The department was unified,and then the target was associated and matched;finally,the early warning system model based on the collision time was designed,the purpose is to verify the above-mentioned perception system,and to discuss the whole system.In the fourth chapter,the software and hardware system of the unmanned vehicle is designed in detail using the above theory,and the actual vehicle system is built.First,the hardware platform and its supporting equipment were selected,and the communication mode was designed.Second,the vehicle software architecture was designed based on the ROS system.Finally,the unmanned vehicle system was designed in modularization and separately The sensor module,perception module,and decision module involved in the subsystems are discussed in detail,and the input and output data interfaces of each module are given.In the fifth Chapter,the content described in Chapter 2 and Chapter 3 are simulated and verified using Chapter 4 as the experimental carrier.Based on C++ programming,the designed perception system is simulated and verified in Gazebo.Finally,based on the data collected by the actual unmanned vehicle,the collision avoidance early warning sub-module is verified by the actual vehicle,which verifies the correctness and effectiveness of the perception system designed in this paper.The sixth Chapter summarizes the above-mentioned full text,and puts forward the shortcomings and improvement directions of this article.