Md Shadab Alam
PhD researcher at TU Eindhoven · Incoming AI Engineer at KPMG Netherlands (from Jan 2027)
I work on computer vision, deep learning and reinforcement learning, and on turning research into systems that hold up in the real world: from studying how people behave in traffic and making automated vehicles safer to building AI solutions in industry.
About me
Full bio, journey & awards →I study how pedestrians and other road users behave in real-world traffic, using computer vision, deep learning and reinforcement learning on large-scale video from around the world. My PhD is in the Department of Industrial Design at TU Eindhoven, and from January 2027 I will join KPMG Netherlands as an AI Engineer, bringing these methods from research into real-world products.
Before that I did an MSc in Robotics and AI at IIT Madras, training deep reinforcement learning controllers for autonomous ships.
Toolbox
Python
C++
C
RustComputer vision
PyTorch
TensorFlowTransformersGraph neural networksReinforcement learning
AWS
Google Cloud
Azure
DockerMLflowDVC
News
All news →- Jan 2027Joining KPMG Netherlands as an AI Engineer from 1 January 2027, alongside finishing my PhD at TU Eindhoven.
- Oct 2026Paper at the UbiComp 2026 companion on what public traffic cameras can reveal about privacy.
- Sep 2026Three papers at AutoUI 2026 in Gothenburg, on generating traffic scenes with Veo 3, personalised EV sounds with generative AI and learning a directional eHMI in VR.
- 2026Our survey of day–night illumination domain translation was published in Machine Vision and Applications.
- Feb 2026Received the FAIR Data Fund grant from 4TU.NIRICT.
What I work on
All research →Pedestrians across the planet
Thousands of hours of YouTube dashcam footage from 238 countries and territories show how people cross streets around the world, well beyond WEIRD samples.
Virtual reality experiments
Pedestrians stand at a virtual kerb as automated cars approach, so we can safely test how yielding, eHMIs, vehicle sounds and other pedestrians change when people feel safe to cross.
AI in place of participants?
Can vision-language models such as GPT-4V predict how people judge risk around automated vehicles and their external displays (eHMIs)?
Generative AI for traffic scenes
Generating realistic traffic scenarios and changing light and weather in traffic video with GANs and video models such as Veo 3.
Deep RL for autonomous ships
Fast-tracking deep reinforcement learning controllers for vessels by first learning from trusted PID and line-of-sight autopilots.
CROWD: City Road Observations With Dashcams
How do pedestrians cross the road in Lagos, Hanoi or Eindhoven? A major part of my PhD is CROWD, a global dataset of dashcam and walking footage collected from YouTube and analysed with computer vision to compare road user behaviour across countries and cultures.
A modular pipeline combines detection, multi-object tracking, semantic understanding, depth estimation and 3D scene reasoning, so we can study pedestrians in natural traffic and support safer automated vehicles in every region.
Spin the globe, hover over a place, or press Take me somewhere to watch real footage from a random city.
Latest papers
All 19 publications →
Personalised Electric Vehicle Acoustics with Generative AI: Dynamic Sonification and User Acceptance Study
18th International Conference on Automotive User Interfaces and Interactive Vehicular A...
Learning a simple directional eHMI: A VR study of AV-pedestrian interaction
Adjunct Proceedings of the 18th International Conference on Automotive User Interfaces ...
Exploring Veo 3's Capabilities for Generating Urban Traffic Scenes in 76 Cities Worldwide
18th International Conference on Automotive User Interfaces and Interactive Vehicular A...
What I'm working on now
11 papers in progress →- Understanding global pedestrian behaviour in 565 cities with dashcam videos on YouTube
- Deep Reinforcement Learning based Eye Tracking for Unity Environment
- Nineteen years of ASMR on YouTube: A multilingual, theme-level analysis of 89,241 videos
- A global dataset of continuous urban dashcam driving
- Collision Patterns and Reporting Blind Spots in 971 California Autonomous Vehicle Crash Reports
- Automated Construction of Living Mobility Datasets from Public Video: A YouTube Case Study
- Understanding Jaywalking Behaviour Worldwide Using Naturalistic Dashcam Videos
- Fixed or randomised trial order? A case study of pedestrian interaction with an automated vehicle
- Perceived Crossing Risk in Multipedestrian Encounters with Automated Vehicles: Effects of Yielding, eHMIs, and Pedestrian Position
- TraffCOCO: A Scalable Framework and Dataset for Traffic Scene Object Detection
- Enhancing driver experience in SAE level 3 automated vehicles through multimodal and emotion-aware in-vehicle agents