Department of Computational and Data Sciences
Department Seminar
Speaker : Prof. Sukhendu Das, Dept. of Computer Science and Engg., IIT Madras
Title : “Catch Me if You Can: A Novel Task for the Detection of Covert Geo-Locations (CGL) from a single image view.”
Date & Time : April 04, 2025 (Friday), 04:00 PM
Venue : # 102, CDS Seminar Hall
ABSTRACT
Most of the visual scene understanding tasks in the field of computer vision involve identification of the objects present in the scene. Image regions like hideouts, corners, bends, turns, and other obscured regions of the scene also contain crucial information for specific surveillance tasks. In this work, we provide an intelligent visual aid for identification of such locations in an image, which either pose an imminent threat or act as target zones for further investigation to identify any concealed objects.
Covert places for hiding behind an occluding object, are concealed locations which are not usually detectable from the viewpoint (camera). Identification of such regions would require knowledge about the 3D boundaries of obscuring items (like pillars, doors, large furniture, wall bends, endings etc.), spatial location and position with respect to the background and other neighboring regions of the scene. To advance towards the goal of identification of such regions, we propose a novel task termed Covert GeoLocation (CGL) Detection. CGL detection finds applications in military counter-insurgency operations and intelligent scene surveillance with path planning for a robot. Given an input RGB image, the goal is to identify all CGLs (potential hideouts) in that image. Since it is not possible to classify any region of an image as a hideout (CGL) without looking at its surroundings, CGL detection would require context-aware detection and understanding of the complex 3D spatial relationships between boundaries of occluding objects and their surroundings. We also highlight the importance of extracting depth information for CGL detection, which the traditional (object) detection approaches often do not consider. Our proposed method successfully extracts relevant depth features from only a single RGB image as input and quantitatively yields significant improvement over existing object detection and segmentation models adapted and trained for CGL detection. We also introduce a novel hand-annotated CGL detection dataset containing ~1.5K real-world images depicting CGLs in diverse indoor environments.
Index Terms: CGL detection, hideouts, location detection, depth perception, visual scene understanding, deep learning.
BIO: Prof. Sukhendu Das is currently employed as a Professor in the Dept. of Computer Science and Engg., IIT Madras, Chennai, India. He completed his B.Tech degree from IIT Kharagpur in the Deptt. Of Electrical Engg. in 1985 and M. Tech Degree in the area of Computer Technology from IIT Delhi in 1987. He then obtained his Ph.D degree from IIT Kharagpur in 1993. His current areas of research interests are: Visual Perception, Computer Vision: Digital Image Processing and Pattern Recognition, Computer Graphics, Machine Learning and Computational Brain modeling.
Prof. Sukhendu Das has been a faculty of the Deptt. of CS&E, IIT Madras, INDIA since 1989, and awarded the CSR changemaker award by IIT Madras in 2023. He has guided about 125 students. He has published more than 200 technical papers (with ~2200 citations – src: google scholar) in international and national journals and conferences. He has completed about 25 Sponsored projects and consultancy assignments. He is a senior member of IEEE, and has received six (6) best papers and a best design contest award. Three of his MS students have recently received best MS thesis awards in the Deptt. of CS&E, IIT Madras.
Host Faculty: Prof. R. Venkatesh Babu
ALL ARE WELCOME
