OPPCON:An Accurate, Efficient Algorithm for Dynamic Feature Extraction
Feifan Zha · 2024
Over the past decades, many impressive SLAM systems have been developed and achieved good performance. However, the strong assumption of scene rigidity limits the use of most Visual SLAM in relevant applications. In this paper, we present the OPPCON algorithm, a dynamic feature detection algorithm in the dynamic environments through curl the optical flow field and find the distance from each feature point to the optical flow plane. Addressed the issue of previous methods not adequately considering the inter-feature constraints within the current frame. We apply our algorithm to ORB-SLAM2 and conducted experiments on challenge dataset such as the TUM RGB-D dataset. The SLAM system with OPPCON methods outperforms the accuracy of standard visual SLAM baselines in high dynamic scenarios. The code will be published at: https://github.com/feifanzha/OPPCON.