A Combination of 3D Faster R-CNN and Laser Scan for a Robust Distance Measurement

Abstract

Measuring the distance of a specific object accurately is a challenging task. Recently, in mobile robot commu-nity, there are several works combining an image-based object detection method and the distance measurement using laser scanner to achieve such challenge. However, the performance of existing methods tend to degrade due to the influence of occlusion and walls behind the target object. To tackle the vulnerability in the existing methods, in this paper, we propose Belief Weighted Max Cluster Average Detection (BWMCAD) method which utilizes the result of 3D Faster R-CNN as the belief of laser scan data and clusters the belief weighted data to extract the distance to the target object. We demonstrate its effectiveness through distance measurement tasks using Tsukuba challenge 2017 data and in-house data in comparison with existing methods.