Eyes on the Road: A Naturalistic Comparison of MTW Rider Gaze in Urban Indian Traffic

Prerak Srivastava; Bhaiya Vaibhaw Kumar; Kavita Vemuri

cs.CVSep 27, 2026

Eyes on the Road: A Naturalistic Comparison of MTW Rider Gaze in Urban Indian Traffic

Authors: Prerak Srivastava, Bhaiya Vaibhaw Kumar, Kavita Vemuri

Organizations: IIIT Hyderabad Hyderabad, India

Abstract

Motorized two-wheelers (MTW) dominate Indian roads but remain underrepresented in driver behavior research. This study presents the first large-scale analysis of MTW driver gaze behavior in naturalistic, heterogeneous urban traffic, using the \textit{myEye2Wheeler} dataset. A semantic segmentation pipeline (YOLOv11 + SAM2) was used to extract object-level gaze metrics under two attention modes: direct gaze (foveal overlap) and central vision (parafoveal monitoring). Results reveal a functional division: central vision supports broad monitoring, while direct gaze enables brief, selective sampling. Novice riders exhibit road-anchored scanning, returning to the road between object fixations, while experienced riders form longer chains of attention across multiple objects. The findings suggest that experience primarily refines temporal rhythm rather than altering allocation strategy and reduces object-class effects in gaze patterns. These findings offer new insight into MTW attention structures and inform future work on behavior modeling and safety systems.

Figures & tables

Panel of Figure 1 — Fig. 1: Example input frame with gaze location ( 1a ) and corresponding semantic segmentation output with detected objects, road mask, and overlaid central vision region ( 1b ).

Direct Gaze
Object	#Obj	%Gaze	P(Gaze)
Car	1125 / 562	8.5 / 7.6	20.4 / 17.6
Motorcycle	528 / 369	3.1 / 2.5	16.1 / 7.8
Motorcyclist	615 / 283	4.8 / 2.4	21.5 / 13.6
M/C (WA)	153 / 87	7.6 / 4.6	20.4 / 16.4
Person	293 / 161	1.2 / 0.9	7.0 / 4.4

TABLE I: GAZE ALLOCATION SUMMARY (After Filtering Outliers)

Panel of Figure 3 — Fig. 2: Gaze transition matrices showing ( 2a ) overall probabilities across all drivers (darker cells indicate higher transition likelihood) and ( 2b ) experience-based differences (warm colors: higher in experienced; cool colors: higher in novice riders).

Direct Gaze
Object	Coverage (%)	Episode Rate (s -1 )	Longest Episode (%)	Fragmentation
Car	31.0(13.7) / 34.0(14.6)	1.6(1.4) / 1.4(1.3)	26.5(27.3) / 28.5(29.2)	0.28(0.24) / 0.22(0.21)
Motorcycle	26.4(12.1) / 34.2(14.5)	1.7(1.5) / 1.7(1.5)	21.5(22.4) / 28.9(27.9)	0.33(0.26) / 0.27(0.24)
Motorcyclist	26.6(11.6) / 24.9(11.5)	1.4(1.2) / 1.1(1.0)	21.3(22.5) / 20.4(22.6)	0.29(0.25) / 0.24(0.19)
M/C (WA)	26.9(9.3) / 28.9(10.9)	1.3(1.1) / 1.0(0.7)	16.8(16.9) / 20.3(20.7)	0.24(0.19) / 0.18(0.12)
Person	40.8(16.2) / 41.8(16.7)	2.4(1.7) / 2.0(1.6)	37.3(32.0) / 38.2(33.4)	0.34(0.28) / 0.30(0.28)

TABLE II: Tracked object attention metrics summary (after outlier filtering)

Explore similar work

MOTOR: A Multimodal Dataset for Two-Wheeler Rider Behavior Understanding
May 21, 2026Varun A. Paturkar, Shankar Gangisetty, C. V. JawaharNaturalistic Driving Data Multimodal Dataset
CVIT, IIIT-Hyderabad, India
Can Webcam Gaze Constrain Mesa-Objectives in Driving Models? An Instrument Precision Analysis
Aug 9, 2026Lennox Anderson, Ahmed Boutar, Jonah Mulcrone +1Gaze Target Estimation Eye Tracking
Duke University, Durham, NC · ONYX AI LLC
From Laboratory to Road: Evaluating Wearable Gaze Accuracy for Driving
Oct 6, 2026William Engel, Fabian Flohr
Intelligent Vehicles Lab, Hochschule München, Germany

May 21, 2026, cs.CVJ/K move · Enter open · S save

MOTOR: A Multimodal Dataset for Two-Wheeler Rider Behavior Understanding

Varun A. Paturkar, Shankar Gangisetty, C. V. Jawahar

CVIT, IIIT-Hyderabad, India

Naturalistic Driving Data Multimodal Dataset Human Activity Recognition Agent Behavior Modeling

Open paper AskPDF arXiv

Two-wheelers account for a disproportionately high share of road fatalities in the Global South. Research on two-wheeler rider behavior, however, lags far behind four-wheelers, where multimodal datasets have driven major advances in Advanced Driver Assistance Systems (ADAS). To address this gap, we present the MOtorized TwO-wheeler Rider (MOTOR) dataset, the first large-scale, multi-view, multimodal resource dedicated to two-wheelers in dense, unstructured traffic. MOTOR comprises 1,629 sequences (25+ hours of video data) collected from 16 riders and integrates synchronized front, rear, and helmet videos, rider eye-gaze from wearable trackers, on-road audio, and telemetry (GPS, accelerometer, gyroscope). Rich annotations capture traffic context, rider state, 12 riding maneuvers spanning conventional and unconventional behaviors, and legality labels (Legal, Illegal, Unspecified). We benchmark rider behavior recognition and maneuver legality classification using state-of-the-art video action recognition backbones (CNN and Transformer-based), extended with multimodal fusion, and find that combining RGB, gaze, and telemetry consistently yields the best performance. MOTOR thus provides a unique foundation for advancing safety-critical understanding of two-wheeler riding. It offers the research community a benchmark to develop and evaluate models for behavior analysis, legality-aware prediction, and intelligent transportation systems. Dataset and code is available at https: //varuniiith.github.io/MOTOR-Dataset/

Abstract

Explore similar work

MOTOR: A Multimodal Dataset for Two-Wheeler Rider Behavior Understanding

Can Webcam Gaze Constrain Mesa-Objectives in Driving Models? An Instrument Precision Analysis

From Laboratory to Road: Evaluating Wearable Gaze Accuracy for Driving