Application and optimization of lightweight visual SLAM in dynamic industrial environment

Abstract

With the increasing adoption of visual SLAM in industrial automation, maintaining real-time performance and robustness in dynamic environments presents a significant challenge. Traditional SLAM systems often struggle with interference from moving objects and real-time processing on resource-constrained devices, resulting in accuracy issues. This paper introduces a lightweight object detection algorithm that employs spatial-channel decoupling for efficient removal of dynamic objects. It utilizes Region-Adaptive Deformable Convolution (RAD-Conv) to minimize computational complexity and incorporates a lightweight Convolutional Neural Network(CNN) architecture to enhance real-time performance and accuracy. Additionally, a novel loop closure detection method improves localization accuracy by mitigating cumulative errors. Experimental results demonstrate the system’s exceptional real-time performance, accuracy, and robustness in complex industrial scenarios, providing a promising solution for visual SLAM in industrial automation.