Current projects span multimodal perception, human-aware assistance, trustworthy AI, and deployable / edge systems for intelligent transportation, with a focus on Intelligent Mobility.
Evaluation of lightweight vision models beyond clean-set accuracy, combining robustness under deterministic corruptions, edge efficiency, explainability, external generalization, and non-compensatory deployment criteria. The work builds on the HCBF preprint and is being substantially expanded for journal submission.
Human-aware supervision for Level 2 automated driving using pre-event gaze and fixation dynamics, driver-grouped conformal risk control, and a graded HUMAN–ASSIST–FALLBACK policy. The study evaluates predictive reliability and downstream safety through counterfactual longitudinal simulation.
An auditing framework for pedestrian detection that stratifies explanations by detector confidence and evaluates explanation behavior under domain shift. The study combines confidence-controlled sampling with perturbation- and activation-based XAI to examine whether explanations remain informative across operating regimes.
A practical Python and OpenCV viewer for real-time RGB, infrared, and depth streams from the Orbbec Astra Stereo S U3. The repository includes interactive exposure, gain, depth-range, and LDP controls, live performance metrics, synchronized frame capture, and a reproducible Windows setup and troubleshooting guide. It provides a practical sensing base for ongoing multimodal (RGB + IR + depth) perception research in intelligent transportation and mobility.
These earlier systems provide the experimental foundation for the broader current agenda in human-aware intelligent vehicles, multimodal perception, trustworthy AI, and adaptive assistance.
CNN evaluation on the public NITYMED dataset for eye-state classification under controlled conditions.
Validation under variable illumination and head motion, emphasizing real-time behavior outside controlled acquisition.
Real-time eye analysis and embedded inference on NVIDIA Jetson Nano for onboard driver monitoring.
Mouth Aspect Ratio analysis integrated with facial landmark tracking as a complementary fatigue cue.