Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

AURASeg: Attention Guided Upsampling with Residual Boundary-Assistive Refinement for Drivable-Area Segmentation

Narendhiran Vijayakumar, Sridevi. M

TL;DR

AURASeg tackles drivable-area segmentation for autonomous robots by emphasizing border refinement and multi-scale context. It introduces an encoder–decoder design with an Attention Progressive Upsampling Decoder (APUD), a Residual Boundary Refinement Module (RBRM), and an ASPP-Lite module atop a CSPDarknet backbone to balance accuracy and real-time performance. Empirical results on outdoor (GMRP) and indoor (Gazebo) datasets show improvements in $mIoU$ and $F1$ over a strong baseline (YOLOP) while achieving high precision with modest speed trade-offs. Overall, the work advances boundary-aware free-space perception for robust navigation in structured and unstructured environments.

Abstract

Free space ground segmentation is essential to navigate robots and autonomous vehicles, recognize drivable zones, and traverse efficiently. Fine-grained features remain challenging for existing segmentation models, particularly for robots in indoor and structured environments. These difficulties arise from ineffective multi-scale processing, suboptimal boundary refinement, and limited feature representation. In order to overcome these limitations, we propose Attention-Guided Upsampling with Residual Boundary-Assistive Refinement (AURASeg), a ground-plane semantic segmentation model that maintains high segmentation accuracy while improving border precision. Our method uses CSP-Darknet backbone by adding a Residual Border Refinement Module (RBRM) for accurate edge delineation and an Attention Progressive Upsampling Decoder (APUD) for strong feature integration. We also incorporate a lightweight Atrous Spatial Pyramid Pooling (ASPP-Lite) module to ensure multi-scale context extraction without compromising real-time performance. The proposed model beats benchmark segmentation architectures in mIoU and F1 metrics when tested on the Ground Mobile Robot Perception (GMRP) Dataset and a custom Gazebo indoor dataset. Our approach achieves an improvement in mean Intersection-over-Union (mIoU) of +1.26% and segmentation precision of +1.65% compared to state-of-the-art models. These results show that our technique is feasible for autonomous perception in both indoor and outdoor environments, enabling precise border refinement with minimal effect on inference speed.

AURASeg: Attention Guided Upsampling with Residual Boundary-Assistive Refinement for Drivable-Area Segmentation

TL;DR

AURASeg tackles drivable-area segmentation for autonomous robots by emphasizing border refinement and multi-scale context. It introduces an encoder–decoder design with an Attention Progressive Upsampling Decoder (APUD), a Residual Boundary Refinement Module (RBRM), and an ASPP-Lite module atop a CSPDarknet backbone to balance accuracy and real-time performance. Empirical results on outdoor (GMRP) and indoor (Gazebo) datasets show improvements in and over a strong baseline (YOLOP) while achieving high precision with modest speed trade-offs. Overall, the work advances boundary-aware free-space perception for robust navigation in structured and unstructured environments.

Abstract

Free space ground segmentation is essential to navigate robots and autonomous vehicles, recognize drivable zones, and traverse efficiently. Fine-grained features remain challenging for existing segmentation models, particularly for robots in indoor and structured environments. These difficulties arise from ineffective multi-scale processing, suboptimal boundary refinement, and limited feature representation. In order to overcome these limitations, we propose Attention-Guided Upsampling with Residual Boundary-Assistive Refinement (AURASeg), a ground-plane semantic segmentation model that maintains high segmentation accuracy while improving border precision. Our method uses CSP-Darknet backbone by adding a Residual Border Refinement Module (RBRM) for accurate edge delineation and an Attention Progressive Upsampling Decoder (APUD) for strong feature integration. We also incorporate a lightweight Atrous Spatial Pyramid Pooling (ASPP-Lite) module to ensure multi-scale context extraction without compromising real-time performance. The proposed model beats benchmark segmentation architectures in mIoU and F1 metrics when tested on the Ground Mobile Robot Perception (GMRP) Dataset and a custom Gazebo indoor dataset. Our approach achieves an improvement in mean Intersection-over-Union (mIoU) of +1.26% and segmentation precision of +1.65% compared to state-of-the-art models. These results show that our technique is feasible for autonomous perception in both indoor and outdoor environments, enabling precise border refinement with minimal effect on inference speed.
Paper Structure (21 sections, 2 equations, 5 figures, 4 tables)

This paper contains 21 sections, 2 equations, 5 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Backbone Architectures and Residual Connections
  4. Multi-Scale Processing and Pyramid Pooling Variants
  5. Multi-Scale Processing and Pyramid Pooling Variants
  6. Attention Mechanisms for Feature Refinements
  7. Boundary Refinement and Edge-Aware Segmentations
  8. Proposed Method
  9. Overview of the architecture
  10. CSPDarknet Backbone
  11. ASPP-Lite Module
  12. Attention Progressive Upsampling Decoder (APUD)
  13. Residual Boundary Refinement Module (RBRM)
  14. Multi-Loss and Supervision
  15. Results
  16. ...and 6 more sections

Figures (5)

  • Figure 1: Overview of the proposed free-space drivable area segmentation encoder-decoder network architecture.
  • Figure 2: ASPP-Lite module merges three parallel convolution branches with dilation rates 1, 6, and 12, each.
  • Figure 3: Integration of Squeeze-Excitation and Spatial Attention Modules for each Attention Progressive Upsampling Decoder (APUD) block
  • Figure 4: Residual Boundary Refinement Module (RBRM) secondary encoder-decoder network architecture.
  • Figure 5: Visual ablation study illustrating the incremental impact of each proposed module. From left to right, outputs progress from the encoder to ASPP-Lite, APUD, and RBRM, culminating in the final refined segmentation.