Improvement And Experiment Of Intelligent Driving Environment Perception Algorithm Based On Prescan

With the development of the automotive industry,smart driving vehicles have become a hot spot for research.Environment sensing technology is a prerequisite for the safe driving of intelligent vehicles,and various sensors are an important medium for vehicles to sense their environment.It is difficult for a single sensor to acquire road information adequately,while the real-time performance of multi-sensor fusion is poor.Therefore,this thesis proposes a real-time detection method for dynamic targets in road scenes based on cameras and LIDAR to provide timely and sufficient environmental information for smart driving vehicles.In addition,this thesis presents a virtual modelling of the road scene for its complexity and possible safety issues,and a comprehensive analysis of the algorithms in this thesis through a virtual road scene.The research in this thesis is as follows:(1)Road scene modelling and simulation.The actual campus road scenario is used as the object and the scenario model is built based on Prescan.A joint simulation platform is established through Prescan/Carsim/Simulink,making the simulation closer to the real road environment.The principle of sensors is analyzed,and the virtual sensors are used to collect information about the vehicle’s surroundings,providing data support for further research.(2)Dynamic target detection for road scenes based on improved YOLOv7.An improved algorithm based on the YOLOv7 algorithm is proposed to address the need for intelligent vehicles to meet the problem of real-time and accurate perception.The speed of detection is improved by replacing the backbone network with a lightweight network,Mobile Netv3;the detection capability of the model for small targets is improved by introducing a small target detection layer and a CBAM attention module;and the detection of the model is made more accurate by improving the loss function.The results show that the m AP and FPS of the improved YOLOv7 algorithm are 92.6% and 75,and the frame rate is increased by 23.0%with a slight loss of accuracy compared to YOLOv7,which can well meet the requirements of real-time and accuracy.Realization of dynamic target detection such as pedestrians and vehicles in road scenes.(3)Sensor fusion based multidimensional perception of dynamic targets in road scenes.A multi-dimensional perception algorithm based on data fusion is constructed.The algorithm achieves target detection through images and distance estimation through point clouds,and finally realizes real-time dynamic target detection and ranging.The experiments show that the fusion algorithm can accurately sense targets and the ranging error at different distances is small,which can meet the task of multi-dimensional sensing of dynamic targets in road scenes,and also meet the requirements of detection accuracy and real-time,which improves the safety and reliability of intelligent driving vehicles.(4)Actual vehicle verification analysis,environmental sensing of the main road of the campus by means of a built intelligent driving test platform.Camera and Li DAR data are collected for fusion through a real-time information sensing system.The results show that the improved LIDAR and camera-based environment sensing algorithm in this thesis can effectively perform the task of detecting dynamic targets such as pedestrians and vehicles in road scenes.