Multi-sensor Fusion Based Traffic Element Detection Algorithm

In the past hundred years,the development of the automotive industry has been dominated by mechanical engineers.Under the environment where computer technology and artificial intelligence are completely infiltrated,the intelligent revolution of automobiles has begun.In order to ensure the safety and reliability of unmanned driving systems,Unmanned vehicles are often equipped with multiple cameras and radar sensors.Which will multiply the burden of the target detection algorithm on the computing and storage units.Therefore,designing high-speed and reliable data fusion algorithms,fully utilizing the advantages of multi-source heterogeneous environment sensing systems will gradually become the research focus in the field of unmanned target detection.This paper proposes a Multi-Sensor-Fusion framework with fault diagnosis and avoidance mechanism,whcih is used to integrate the environmental perception data of camera and radar sensor to complete the 3D target detection task.Compared with traditional Multi-Sensor-Fusion methods,the main innovations of this paper are as follows:· A method of fusing sensor data from multiple cameras is proposed.It maps the2 D ROIs predicted by different cameras to the 3D space to form the cone region,and design the corresponding algorithm to fuse those cone.· The fault diagnosis and avoidance mechanism is introduced into the sensor fusion framework,which ensures the accuracy and reliability of environmental perception.Finally,this paper tests the performance of the data fusion network from three aspects:the accuracy of 2D target detection,the accuracy of 3D target detection,and the ability to avoid sensor failures.The experiments prove that the fault diagnosis and avoidance mechanism in this paper can effectively detect and avoid global and local sensor failures in a variety of scenarios;The lightweight sensor fusion network greatly reduces the computing power and memory space required by the network while ensuring the accuracy of target detection,which has a certain engineering application value.