Prerak Srivastava supervised by Dr. Kavita Vemuri  received his Master of Science – Dual Degree  in Computer Science and Engineering (CSD). Here’s a summary of his research work on  Visual Attention Patterns of Indian MTW Drivers in Naturalistic Complex Heterogeneous Traffic: A Statistical Analysis of Novice and Experienced Riders: 

Motorized two-wheelers (MTWs) dominate Indian roads but remain underrepresented in driver behavior research, particularly regarding gaze behavior in naturalistic, heterogeneous urban traffic. This study analyzes MTW rider visual attention using the \textit{myEye2Wheeler} eye-tracking dataset, comprising 217,012 frames from 25 riders (15 experienced, 10 novice) navigating a 3.9 km urban route in Hyderabad, India. A computer vision pipeline combining YOLOv11 for object detection, ByteTrack for multi-object tracking, and SAM2 for instance segmentation extracts object-level scene representations, which are then linked with frame-level gaze coordinates to produce attention metrics under two modes: direct gaze (foveal overlap with object masks) and central vision (parafoveal monitoring within a 13$^\circ{}$ elliptical zone). Results establish a functional division between the two attention mechanisms. Central vision operates as a sustained monitoring system, covering a median 62.5\% of tracked object lifespans, while direct gaze functions as a selective sampling mechanism, covering only 12.3\%. This division persists across all object classes and experience levels. Spatial analysis reveals that novice riders concentrate 56.8\% of fixations in the central visual field compared to 41.3\% for experienced riders, who redistribute attention toward the upper field for distant traffic monitoring. Gaze transition analysis shows novices employ road-anchored sequential scanning, returning gaze to the road surface after 66\% of object fixations, whereas experienced riders form distributed multi-object attention chains with 32\% direct object-to-object transitions. Experience effects manifest primarily in central vision, where experienced riders show reduced coverage, higher episode rates, and greater fragmentation, indicating a shift from sustained peripheral tracking to efficient intermittent sampling. Direct gaze patterns remain largely unchanged across experience levels, suggesting foveal deployment follows universal task demands. Object-class effects on attention progressively flatten with experience, transitioning from salience-driven allocation in novices to near class-invariant strategic monitoring in experienced riders. 

May 2026