Guiding Robotic Cloth Grasping in Darkness Infrared Semantic Segmentation and Grasping Position Selection

Project Code :TCMAPY2302

Objective

"To implement and compare the performance of UNet++, Swin-UNet, and DeepLabV3+ models for semantic segmentation of cloth in infrared images.To design an algorithm that analyzes segmentation masks to select optimal coordinates for robotic grasping. To develop a functional web application with user authentication for uploading images and visualizing segmentation and grasp point results.To integrate the trained segmentation model and grasping algorithm into the web application backend using Flask. To document the complete system workflow, from data preprocessing and model training to user interaction, providing a B46:H46reproducible research framework"

Abstract

This research presents an autonomous system enabling robotic cloth grasping in darkness by integrating infrared vision, semantic segmentation, and grasp point computation. To solve the visual challenge, the project trains and evaluates multiple encoder-decoder deep learning models—specifically UNet++, Swin-UNet, and DeepLabV3+ with a MobileNet backbone—on a dataset of infrared images and corresponding pixel-level mask annotations. The selected optimal model performs robust semantic segmentation to accurately delineate cloth objects from the background in infrared scenes. The resulting segmentation mask is algorithmically analyzed to identify coordinates for stable and effective robotic grasps based on cloth geometry and inferred physical properties. For accessibility and demonstration, the complete pipeline is deployed via an interactive web application built with standard front-end technologies (HTML, CSS, JavaScript), where users can upload an infrared image and receive the segmented result alongside the proposed grasp point visually overlaid on the output.

Keywords: Robotic Manipulation; Cloth Grasping; Infrared Imaging; Semantic Segmentation; Deep Learning; UNet++; Swin-UNet; DeepLabV3+; Grasp Point Detection; Web Application.

NOTE: Without the concern of our team, please don't submit to the college. This Abstract varies based on student requirements.