Space-variant Color Point Cloud Measurement System - Enomous Data Reduction using Saliency Map -
Sota Shimizu, Yu Fujita, Naoaki Kameyama, N. Hasebe · 2020
This paper aims at reducing enormous data amount of color point cloud measured from a 3D-LIDAR sensor by combining with a visible-ray color camera. A huge data amount of the points often keep us away from saving the data storage device and computing by a real-time processing. In this paper, the authors propose a measurement system, i.e., Space-variant Color Point Cloud Measurement System, in order to achieve enormous data reduction. A 2D-LIDAR sensor and a stepping motor are combined (i.e., playing a role of the 3D-LIDAR sensor) to compose a proto-type of our experimental device. A space-variant color point cloud generation method is also proposed using the above measurement system. The region of interest (ROI) map, generated from the saliency map of the input image from the CMOS color camera, plays an important role for the data reduction in our proposed method. This paper explains also how a pixel with RGB color data in each input image corresponds to each of the distance data, by paying attention to a disparity calculated from the distance. Experimental results proved good performance of data reduction using our proposed method by 60.59 percent and 40.82 percent in different reduction conditions compared to the original data amount, respectively.