Similarity Fusion: Camera-LiDAR Vision Similarity Fusion for 3D Object Detection
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
This paper proposes a novel vision similarity fusion network for fusing 2D camera images and 3D LiDAR point clouds, utilizing merged vision feature for 3D object detection. Similarity fusion network provide a non-coupled, low-complexity vision fusion framework that significantly improves the ability to connect heterogeneous vision sensor. Our fusion network fuses the proposed candidate combinations of camera and LiDAR decoders before non-maximum suppression. A vision correlation fusion law is designed using Cosine similarity, and non-coupled vision features is fused according to correlation without additional training. Finally, we designed a fusion detection head to achieve 3D detection of the object using a combination of visual features. Experiments on the challenging KITTI object detection library show that our fusion scheme significantly improves object detection, especially for small objects, compared to individual point cloud and image detection schemes. We will release our code after acceptance.
Fields of Study & Subjects
Advanced Neural Network Applications
Autonomous Vehicle Technology and Safety
Advanced Vision and Imaging
Publication details
- Journal / source
- CRC Press eBooks
- Pages
- 46–57
- Publication date
- 27 March 2026
- Language
- EN
- Document type
- Book chapter