Research On Environment Perception Algorithm Of Unmanned Vehicle Based On Multi-source Sensor Fusion

With the development of intelligent technology and the increase of market applications,unmanned vehicles gradually appear in our life scene.As one of the key technologies of unmanned vehicles,environment awareness technology directly or indirectly affects the intelligent level of vehicles,which is a research hot spot in the field of unmanned vehicles.Complete and accurate environmental perception information is the data basis for the safe and stable operation of unmanned vehicles,and its importance can not be ignored.The realization of environment perception function of unmanned vehicle includes two parts: environment information collection and perception recognition.The realization of vehicle environment perception function is based on the collection of environment information around the vehicle by sensors.The accuracy and integrity of the collected data directly affect the accuracy of subsequent perception recognition,while single sensor data collection often has environmental robustness Low speed,poor stability,obvious detection blind area and other shortcomings,which seriously affect the driving safety of unmanned vehicles.Based on the perception needs of urban passenger cars,this paper studies an environment perception algorithm of unmanned vehicles based on multi-source sensor fusion,aiming to effectively improve the accuracy and real-time of environment perception of unmanned vehicles,and strive to avoid the security risks caused by perception errors or incomplete.The environment sensing algorithm of unmanned vehicle in this paper includes two parts: Multi-source sensor fusion strategy and environment sensing algorithmMulti sensor information fusion can effectively make up for the single sensor detection blind area,vulnerable to the surrounding driving environment and other problems,so as to provide reliable and complete driving environment information for the follow-up unmanned vehicle environment perception,which is the data cornerstone and safety guarantee of the whole unmanned driving system.So in this paper,based on multi-sensor fusion,we further study the environment sensing algorithm for unmanned vehicles.Due to the high requirement of real-time processing for unmanned vehicles,the existing semantic segmentation algorithm has high computational complexity and low network inference speed.For this kind of task,this paper focuses on improving the problem of "contradiction between accuracy and real-time" existing in the existing perception algorithm,aiming to find the best balance between accuracy and real-time In addition,we use multi-sensor data as input to obtain the semantic categories of driving environment and the motion information of obstacles,so that the perception algorithm can really go to the road of application.Based on the computational complexity formula,this paper analyzes and studies the existing segmentation algorithms,and proposes a high real-time semantic segmentation model based on attention mechanism,which can segment the Cityscapes dataset in real time.Finally,we test our network on two large semantic segmentation datasets.Our network achieves real-time performance on Cityscapes dataset,and achieves leading accuracy,such as 71 FPS / 79.9% m Io U,130 FPS / 78.5% m Io U and 180 FPS / 70.1%m Io U.In addition,our performance on ADE20 K is also in a leading position.