📝 SummarXiv

Abstract

With the rapid development of lightweight visual neural network architectures, traditional high-performance vision models have undergone significant compression, enhancing their computational and energy efficiency and enabling deployment on resource-constrained edge devices. In order to enable the mobile robot to avoid the ground wires, we propose a visual-assisted navigation framework called Atrous Strip Convolution Sliding Window (ASC-SW). This framework compensates for the limitations of traditional light detection and range (LiDAR) sensors to detect ground-level obstacles such as wires. A lightweight and efficient segmentation model, Atrous Strip Convolution Network (ASCnet) was proposed, for detecting deformable linear objects (DLOs). Atrous Strip Convolution Spatial Pyramid Pooling (ASCSPP) is designed to extract DLOs features effectively. Atrous Strip Convolution is integrated into ASCSPP to accurately identify the linear structure of DLOs with low computational cost. Additionally, a Sliding Window (SW) post processing module is proposed to denoise the output in complex environments, improving recognition accuracy. ASC-SW achieves 75.3% MIoU at 217 FPS on a self-built real world dataset and real-robot experiment was demonstrated that our proposed framework. It can be successfully verified on the real-robot on the edge device(Jetson platform) at that were originally inoperable.