Skip to main content

Hwang’s team wins competition in AI challenges

July 3, 2018

Every day, people are using artificial intelligence (AI) in a variety of capacities to make lives easier, safer and more advanced—including in transportation. A team of UW graduate and undergraduate students and researchers has been recognized for its significant research in making transportation systems smarter and safer.

The group, led by Electrical Engineering Professor Jenq-Neng Hwang, won two out of three track challenges from the AI City Challenge at the 2018 IEEE Computer Vision and Pattern Recognition (CVPR) held in Salt Lake City. Graduate students Zheng Tang, Gaoang Wang and Hao Xiao, and undergraduate student Aotian Zheng, researchers in Hwang’s Information Processing Lab, were a part of the award-winning team.

Hwang and his team, Zheng Tang, Gaoang Wang and Hao Xiao, and undergraduate student Aotian Zheng, receiving their award.

Hwang and his team, Zheng Tang, Gaoang Wang and Hao Xiao, and undergraduate student Aotian Zheng, receiving their award.

According to Hwang, The AI City Challenge tasks competitors to interpret copious hours of traffic footage from cameras on the highways and streets to better understand transportation. Each track challenge has footage from different locations. Teams have to use their own programs to analyze the footage and find the solutions to problems given to them for the challenge.

In the first track, teams were given a test set of 27 one-minute videos and told to analyze the traffic flow. They were told to analyze the speed of the cars in the video. There were “control” cars in the videos, too. These cars were placed in traffic by the challenge, so that they knew the actual speed of some of the cars in the video. Of the 56 teams participating in this challenge, only 13 teams eventually submitted their analyzation, with Hwang’s team doing the best of all of the competitors.

Track two, which the UW team did not participate in, was about anomaly detection. Groups were supposed to analyze 100 detected anomalies and find out what caused car crashes or stalled vehicles. And the third track challenge was about using multi-sensor vehicle detection and re-identification to see which vehicles passed four different locations in a set of 15 videos. While each team used different AI tech to interpret the data, Hwang’s team once again won track three, beating out 61 teams that tried to participate and 10 that actually submitted results of the challenge.

According to Hwang, the teams have published their results and submitted the open-source codes they used so that the information is public for everyone to improve upon for the common good. Additionally, the group won two advanced graphics processing units (GPUs) from the challenge sponsor, NVidia company.

“We also get a lot of visibility,” said Hwang about their wins. “In our case, there were already some companies approaching us at the conference. It’s important technology for security, smart cities, customer service and even autonomous driving.”