The Design And Implementation Of Autopilot Platform And Lane Detection Algorithm Based On Model Vehicle

The rapid development of computer technology has greatly promoted the research of the artificial intelligence field.With the breakthroughs in electric vehicles,sensors,and other related areas,autonomous driving has become one of the most active topics.Although autonomous driving technology is now under a pivotal period,the extremely high research thresholds and road test costs have become bottlenecks restricting its further development and promotion.Based on the facts above,an open,lightweight,and high control accuracy platform for better research of autonomous driving is designed and built.Moreover,a lane detection module is proposed and implemented based on the designed platform.The main contents are as follows:Firstly,based on the general structure of autonomous driving systems,an architecture for the autonomous driving platform that accurately reflects the characteristics of real vehicles is designed.And the application layer was modularized following the "perception-decision-manipulation" behavior model.An effective and feasible implementation scheme is proposed according to the platform architecture.In detail,a model car with a scale of 1:10 to the real vehicle is selected as the platform chassis,and the hardware modules are integrated into the chassis to comply with the hardware platform integration.The open-source ROS middleware is selected as the development and operation framework to unify the application access method.Through the conceptions above,the physical construction and the integration of the software layer for autonomous driving are completed.Secondly,this paper proposes an instance segmentation algorithm by using gridding method for extracting lane trajectories,which avoids the repeated processing of redundant information in the traditional dense prediction method.The proposed algorithm divides the image into multiple small cells for classification,thereby reducing the redundant information in the extraction process.Moreover,a deep neural network is constructed and then implemented through Python and TensorFlow.The network is compared with the Lane Net on the CULane dataset to verify its performance.The results show that the improved lane detection algorithm has significant advantages in terms of accuracy and real-time performance.Based on the algorithm,a perception module for lane detection is developed to realize the lane perception function in the model car.At last,the autonomous driving platform along with the perception module is tested through multiple experiments.By studying the relationship of the given speed value with the actual speed and given steering angle value with the actual steering angle,the control accuracy of the model car is quantitatively analyzed.Besides,the priority of the related system interface is verified.Through the platform stability experiment,the relationship among the motor input current,the startup speed,and the startup acceleration are investigated.By comparing the recognition speed of the sensing module with Lane Net,the real-time performance of the sensing module is verified in the driving state.Furthermore,the video data of the experimental site is collected,and the data is manually labeled then to become a dataset,on which the accuracy of the perception module is verified.It provides a vital foundation for the further development of decision and control modules.