Reducing Falsely-detected Feature Points of SLAM by Estimating Obstacle-free Area for RCMSs

Abstract

In recent years, wireless sensor network (WSN) based monitoring systems have been applied in agricultural vermin control. For example, there are trap systems that capture vermin by automatically closing their gates. These systems need to monitor vermin that approach farmland. Ropeway camera monitoring systems (RCMSs) have been proposed as vermin monitoring mechanisms. In an RCMS, cameras can move along ropes stretched between trees or poles. However, a problem in RCMSs is that obstacles lead to poor visibility, and cameras cannot monitor areas effectively. Therefore, it is crucial to estimate locations of obstacles such as tree trunks. When estimating locations of obstacles using simultaneous localization and mapping (SLAM), it is difficult to extract feature points in dense vegetation due to noise and brightness issues. As a result, feature points are sometimes falsely detected in locations where there are no obstacles. In order to improve SLAM accuracy, falsely-detected feature points must be identified. In this study, we propose a method to estimate obstacle-free areas for an RCMS. The proposed method can determine falsely-detected feature points in estimated obstacle-free areas, and reduce errors in SLAM. The proposed method determines the largest obstacle-free areas, while reducing the number of camera shots. A camera in an RCMS also shoots other cameras while moving along its rope. When the camera captures the other cameras, we find that there are no obstacles between two cameras. The proposed method effectively identifies obstacle-free areas by moving two cameras simultaneously. From the results of a simulation with two parallel ropes, we confirmed that the proposed method determines approximately 92% of obstacle-free areas, compared with the brute-force algorithm.