Multi-sensor feature fusion for vehicle detection based on the fuzzy longest common subsequence

To mitigate the impact of sensor misdetection and omission on the multi-sensor feature fusion algorithm, this paper proposes a vehicle detection method for multi-sensor feature fusion based on the fuzzy longest common subsequence algorithm. Firstly, the lightweight YOLOv4 algorithm is utilized to obtain image features. Secondly., an adaptive threshold-based clustering method processes point cloud data, extracting trajectory information from both image and point cloud sources. Then, an enhanced longest common subsequence algorithm, incorporating a fuzzy membership function, is introduced to assess the similarity between the lidar target trajectory and the camera target trajectory. Next, the fusion of lidar and camera detection results is carried out based on the computed similarity scores. Ultimately, experimental validation affirms that this algorithm significantly improves vehicle detection accuracy. This enhancement contributes to a more robust and reliable vehicle environment sensing system.