go to contents go to side menu go to main menu

main menu area

main menu area

Innovative University Changing the World through Convergence

Embed to SNS

Research News

DGIST Gains Top International Ranking in Vehicle Classification and Localization

  • 조회. 537
  • 등록일. 2017.08.29
  • 작성자. Administrator

DGIST Gains Top International Ranking in Vehicle Classification and Localization 

- Top rank at TSWC, CVPR 2017 achieved using deep neural network convergence technology - 

DGIST Convergence Research Center for Future Automotive Technology ranked first at an international challenge devoted to vehicle classification and localization.         

DGIST announced on Wednesday, July 26 that its future automotive technology center, a division of the DGIST Convergence Research Institute, entered and attained top ranking in the Traffic Surveillance Workshop & Challenge (TSWC) held in conjunction with Computer Vision and Pattern Recognition 2017 (CVPR 2017), the world’s largest event in the computer vision and pattern recognition discipline. 

The Traffic Surveillance Workshop & Challenge is an arena for competition in the capability of identifying moving vehicles and pedestrians from CCTV images and the participants challenge their algorithms in two parts: Vehicle Classification and Localization. 

For the vehicle classification part, each algorithm will be evaluated with its capability to identify vehicles of all kinds, including bicycles, buses, cars and trucks, as well as pedestrians, from 11 different ranges of datasets; while the vehicle localization challenge is aimed at determining the location of vehicles with maximum accuracy after they had been identified.      

By applying deep learning technology which has recently been receiving great attention in the field of artificial intelligence, the Center accomplished a remarkable achievement by taking first place in both the Vehicle Classification and Localization categories with accuracies of 97.95% and 79.24% respectively.   

The center presented a paper on deep learning-supported vehicle classification and localization technologies during the TSWC at CVPR 2017, held recently in Hawaii, attracting considerable attention from researchers in related disciplines. The technology presented involved new techniques for converging multiple deep neural networks which can be installed on self-driving vehicles; and the record-high results the center achieved in the challenge were set utilizing this technology.    

“Deep learning-based vehicle classification and localization technology is the foundation for intelligence traffic systems and is at the very core of the fourth industrial revolution.” stated Dr. Woo Young Jung, director of the center. “Based on this achievement, the center will continue its research efforts to lead AI self-driving vehicle technology both at home and abroad.”       

Last year, the DGIST Convergence Research Center for Future Automotive Technology also participated in ImageNet Large Scale Visual Recognition Challenge 2016 (ILSVRC 2016), an arena for international competition in artificial intelligence, ranking sixth with a 3.291% error rate in the category of Object Detection.