Scholarlink

<p>This paper presents a real-time 3D LiDAR perception framework, enhanced with an Attention-Driven MobilePIXOR model for obstacle segmentation and pedestrian detection in autonomous robots. The proposed multi-task model leverages the MobilePIXOR architecture with depthwise separable convolutions and integrates Convolutional Block Attention Module (CBAM), enhancing object detection precision to 96.80% and improving IoU for obstacle segmentation to 56.50%. A heatmap header system, incorporating Gaussian heatmaps, is used to predict object center locations, further improving detection accuracy. The framework converts LiDAR data into structured voxel grids, optimizing spatial awareness and enabling the differentiation between drivable and non-drivable areas. An integrated loss function is introduced to jointly optimize both detection and segmentation tasks. Additionally, the novel <InlineEquation ID="IEq1"> <EquationSource Format="TEX">\( Lidar3D\_SEG[NONSPACE] \&amp; DET\)</EquationSource> <EquationSource Format="MATHML"><math> <mrow> <mi>L</mi> <mi>i</mi> <mi>d</mi> <mi>a</mi> <mi>r</mi> <mn>3</mn> <mi>D</mi> <mi>_</mi> <mi>S</mi> <mi>E</mi> <mi>G</mi> <mo stretchy="false">[</mo> <mi>N</mi> <mi>O</mi> <mi>N</mi> <mi>S</mi> <mi>P</mi> <mi>A</mi> <mi>C</mi> <mi>E</mi> <mo stretchy="false">]</mo> <mo>&amp;</mo> <mi>D</mi> <mi>E</mi> <mi>T</mi> </mrow> </math></EquationSource> </InlineEquation> dataset was developed to evaluate the framework’s effectiveness. Experimental results demonstrate that the enhanced MobilePIXOR model with CBAM outperforms other architectures in real-time performance, making it ideal for autonomous navigation in dynamic environments.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Multi-Task Real-Time 3D LiDAR Perception with Attention-Enhanced MobilePIXOR for Obstacle Segmentation and Pedestrian Detection in Autonomous Robots

  • 期刊论文

  • 发表日期：

    2025年9月1日

查看全文
  • Hoang N. Tran,
  • Cong T. Nguyen,
  • Truong T. Duong,
  • Minh L. Nguyen,
  • Vinh D. Nguyen

摘要

This paper presents a real-time 3D LiDAR perception framework, enhanced with an Attention-Driven MobilePIXOR model for obstacle segmentation and pedestrian detection in autonomous robots. The proposed multi-task model leverages the MobilePIXOR architecture with depthwise separable convolutions and integrates Convolutional Block Attention Module (CBAM), enhancing object detection precision to 96.80% and improving IoU for obstacle segmentation to 56.50%. A heatmap header system, incorporating Gaussian heatmaps, is used to predict object center locations, further improving detection accuracy. The framework converts LiDAR data into structured voxel grids, optimizing spatial awareness and enabling the differentiation between drivable and non-drivable areas. An integrated loss function is introduced to jointly optimize both detection and segmentation tasks. Additionally, the novel \( Lidar3D\_SEG[NONSPACE] \& DET\) L i d a r 3 D _ S E G [ N O N S P A C E ] & D E T dataset was developed to evaluate the framework’s effectiveness. Experimental results demonstrate that the enhanced MobilePIXOR model with CBAM outperforms other architectures in real-time performance, making it ideal for autonomous navigation in dynamic environments.