Previous Projects
Scaled Platform for Cooperative Driving Automation.
We developed a 1/10th scale hardware-software platform enabling vehicles to communicate with one another and infrastructure (V2X). This platform facilitates fleet coordination, reduces traffic congestion, and improves overall road safety through shared sensor data and real-time decision-making protocols.
2COOOL: 2nd Workshop on the Challenge Of Out-Of-Label Hazards in Autonomous Driving.
As leaders in the field, we organized the 2nd Workshop on the "Challenge Of Out-Of-Label" hazards at the International Conference on Computer Vision (ICCV). This project focused on identifying edge cases and anomalies in autonomous driving data - ensuring AI systems can handle the unpredictable elements of real-world roads that standard datasets often miss.
Audio Way Finding for Visually Impaired Pedestrians.
Utilizing advanced computer vision and spatial audio, we created a navigation assist tool for the visually impaired. The system identifies obstacles, crosswalks, and safe paths in real-time, converting visual data into distinct audio cues to empower independent and safe mobility.
Autonomous Vehicle-Pedestrian Interaction Modeling Platform
A critical safety initiative focused on predicting human behavior. This platform models the complex social interactions between self-driving cars and pedestrians, using deep learning to anticipate movements and ensure vehicles react naturally and safely in crowded urban environments.
3D Pedestrian Tracking in Large Scale Synthetic Videos via Multi-Camera Fusion.
We tackled the challenge of tracking individuals across large-scale environments using synthetic video data. By fusing inputs from multiple camera angles, this system maintains consistent ID tracking through occlusions and complex scenes, essential for security and safety analytics.
Parallel Dense Video Captioning for end-to-end event analysis.
Designed for end-to-end event analysis, this system automatically generates detailed textual descriptions of video content in parallel. It allows for rapid indexing and searching of video archives, transforming hours of raw footage into searchable, structured data for security or operational insights.
