Automatic Driving 3D Object Detection Based on Lidar and Camera Information Feature Fusion
TL;DR
This review offers an all-round synthesis of how LiDAR and camera sensors are integrated for 3D target detection and aims to give a comprehensive theoretical explanation to scholars who have a preliminary understanding of the fusion of LiDAR and camera.
Abstract
This review offers an all-round synthesis of how LiDAR and camera sensors are integrated for 3D target detection. It first outlines why LiDAR–camera fusion is necessary and where its advantages lie, then turns to the respective characteristics and underlying theoretical bases of LiDAR and cameras, particularly with respect to detection range, measurement accuracy, and semantic richness. From there, the discussion examines the operating principles of both sensing modalities and, more to the point, the central challenge in fusing them: the heterogeneity of the data each system acquires. Against that technical backdrop, the analysis proceeds to the current mainstream approaches to feature-level fusion. Through feature- level fusion, LiDAR point cloud features and camera image features can be effectively integrated, which broadens the environmental perception scope and supports accurate extrinsic calibration for reliable measurement. Finally, this paper further analyzes the existing problems in this field and prospects the future development direction. This paper aims to give a comprehensive theoretical explanation to scholars who have a preliminary understanding of the fusion of LiDAR and camera.