[1Yin-A] Poster 1

[1Yin-A-37] Associating Image Regions with LiDAR Point Clouds Using Point Cloud Cluster Information

〇Kyosuke Watanabe¹, Seiji Amamoto¹, Osamu Tokumi¹ (1. SECOM Co., Ltd.)

Keywords:

Sensor Fusion,Image Processing,LiDAR

To achieve high-precision fusion of multiple sensors, accurate cross-sensor correspondence of target objects is essential. Research and development datasets for training and inference presuppose this correspondence, and algorithms rely on its correctness. Traditionally, correspondence has been established at the smallest unit level (e.g., image pixels and LiDAR points) using external calibration parameters, but projection errors and other factors often cause mismatches. To address this issue, we propose a method that performs correspondence at a higher-level cluster unit. Validating the method with image–LiDAR pairs, we confirmed improved accuracy compared with minimal-unit correspondence.