Estimations assert that there will be 1 billion cameras on the road by 2020. These cameras present a significant opportunity for transportation; their data can offer actionable insights to make transportation systems smarter and safer. However, current systems present challenges. Poor data quality, a lack of labels for the data and the lack of high quality models that can convert the data into actionable insights are some of the biggest blockers to harnessing the data’s value.
The AI City Challenge requested research proposals for two tracks to address these problems. For Track 1, researchers were asked to develop models for basic machine learning tasks as they applied to current transportation systems’ data. Track 2 researchers were asked to propose and develop AI City applications geared towards solving salient problems related to safety and/or congestion in an urban environment.
The team of researchers in Professor Hwang’s Information Processing Lab won the Track 2 challenge for AI City Applications. Their work focused on a constrained multiple-kernel (CMK) tracking system to resolve the problem of occlusion during multiple object tracking. This model also enables researchers to understand vehicle attributes, like vehicle type, speed, orientation, etc. The researchers also proposed several future improvements on their current work, including license plate identification and application in multiple-camera tracking. Their experiments on the NVIDIA dataset outperformed several state-of-the-art algorithms in tracking by segmentation and tracking by detection.
The AI City Challenge is jointly sponsored by IEEE and NVIDIA through the IEEE Smart World Congress annual conference. Researchers for Track 2 included Zheng (Thomas) Tang, Gaoang Wang, Tao Liu, Young-Gun Lee, Adwin Jahn, Xu Liu, Dr. Xiaodong He from Microsoft Research and Professor Hwang.
Team lead Zheng Tang urged that this win was a team effort, thanking the UW undergraduate students, who offered assistance: Lingli Zeng, Aotian Zheng, Yan Kuo, Kevin Nguyen, Jingwen Sun and Chien-Jen Hwang.