| PhD Seminar

Name of the Speaker: Mr. Nambala Ramsai (EE20D056)
Guide: Dr. Sridharan
Venue: ESB-234 (Malaviya Hall)
Date/Time: 25th April 2025 (Friday), 3:00 PM
Title: Object Detection and Distance Measurement in Robotics Under Non-Ideal Conditions

Abstract :

In classical robotic systems, tasks have been performed with the help of sensors mounted on the robot or data from surrounding sensors. These tasks have been accomplished using basic algorithms or task-specific pre-programmed rules and typically, classical vision techniques. For robotic navigation, Simultaneous Localization and Mapping (SLAM) techniques, visual odometry, feature matching, and stereo vision techniques have been employed. Histogram of Oriented Gradients(HOG), colour-based segmentation, and Canny-type edge detection are some of the early methods used for object detection tasks. Although they achieve good performance in near ideal conditions, they have limitations in several practical environments. With the advancement of deep learning techniques and high-end computation hardware, the performance of robotic systems for various tasks has drastically improved. However, learning alone has not typically addressed the problems faced. A combination of sensor fusion with deep learning techniques has contributed to enhanced robot performance catering to various tasks. Deep networks such as CNNs and RNNs have been proposed to handle environment variations, which include distortions, noise and dynamic changes in the environment. However, these networks on their own do not guarantee consistent performance in more challenging scenarios, which include occlusion, variations in illumination, etc. In this seminar, we will consider the role of contemporary learning methods for detecting objects under constraints in the context of mobile robotics. We will give an overview of deep networks with specific focus on robot tasks in a personal care setting. We will then present briefly our work on development of a generative model for object detection and distance measurement under limited illumination.