Video Analytics

In the past decades, naturalistic driving studies (NDSs) have enabled significant advances in the understanding of real-world driver behavior and its relation to crash involvement. Coming to useful conclusions from NDS data, however, most often requires the manual reduction of large quantities of data by humans, which is labor intensive, time consuming, and costly.

As part of my research assistantship at Virginia Tech Transportation Institute, I was mainly involved in this project. My role was to analyze naturalistic driving videos and automatically produce annotations and descriptors for events, behavior, and driving scenarios that relate to transportation safety and operation. My focus was to develop novel and accurate computer vision algorithms and systems, leveraging deep learning and machine learning on big data resources. At the same time I analyzed and improved efficiency, scalability, and stability of various deployed systems for tasks. A demo is not available due to privacy restrictions. (picture from Center for Truck and Bus Safety)