A Multimodal Target Detection Method Based on the MES and the Improved BEV Perception Algorithm

Three-dimensional target detection plays a vital role in the autonomous driving system providing environment perception and the obstacle detection capabilities for its decision-making and control. The paper proposes a multimodal target detection model to improve the BEV (Bird's Eye View) perception. The model adopts a new parallel fusion module to efficiently integrate the laser point cloud and the image information. The paper combines the network with the MIVAR (Multidimensional Informational Variable Adaptive Reality) expert system to improve accuracy of the target detection and decision analysis. The paper creates a knowledge base, analyzes the multi-modal 3D target detection method using a combination of the advanced LiDAR (Light Detection and Ranging) and camera, and designs the corresponding decision-making rules. Verification was carried out on the KITTI (Karlsruhe Institute of Technology and Toyota Technological Institute) dataset, and experimental results showed that the 3D target detection and the decision-making accuracy were significantly improved.