1:10 Scale Autonomous Driving Dataset for Lane Detection and Vehicle Control

This dataset was created to support research on lane detection for 1:10 scale autonomous vehicles (AutomoModelCar). It contains synchronized infrared image sequences and region-of-interest (ROI) annotations collected using an Intel RealSense D435 mounted on a scaled autonomous platform designed for the AutoModel Car and TMR2025 competition. The dataset includes raw 5024 mages, and approximately 4,056 ROI’s manually labeled (line vs. no-line). Each raw frame is timestamped and synchronized with vehicle telemetry CSV column steering, providing both visual and contextual driving information, steering value range from full left turn 1,000 up to 2,000 right turn, where 1,500 is the middle and the car is going straight.

The dataset was curated to simulate realistic small-scale driving scenarios, including variable illumination, shadows, faded lane markings, and curved tracks. It is intended for training and evaluating lightweight convolutional neural networks and other vision-based approaches for lane detection under strict computational constraints.