Deep learning-based object detection models have been applied to various applications over the past years for their far better performance compared to traditional learning methods. However, deep learning-based models have more parameters than traditional ones, require more data to train, and intrinsically rely on computing power for accelerating inference speed. In addition, it is not trivial for deploying deep learning-based systems to a resource-limited device to obtain real-time and high model performance. In this talk, I will share our recent study on developing a deep-learning-based object detection system on a field-programmable gate arrays (FPGAs) platform for driving applications. As an alternative to graphics processing units (GPUs), the developed system makes use of heterogeneous multiprocessing for computation acceleration of deep neural networks to provide low latency and has lower energy consumption compared with GPUs, making it suitable for unmanned aerial vehicles (UAVs) and autonomous driving applications.
Dr. Wen-Hui Chen earned his Ph.D. degree from National Taiwan University (NTU) in 2000. He joined the National Taipei University of Technology (NTUT) in 2002, where he is currently a Professor at the Graduate Institute of Automation Technology. He was a two-time Taiwan Fellowship recipient from the Ministry of Education in 2001 and 2003, and a Research Abroad Award recipient from the Ministry of Science and Technology in 2013. During the 2013 academic year, he was a visiting scholar in the Department of Computer Science, University of Pittsburgh. From 1992 to 2000, he worked as a System Engineer at the Information and Supervisory Control Division, Taiwan Power Company (Taipower Co.), Taiwan's largest power utility company. His research interests lie in the areas of deep learning applications in computer vision and edge AI ranging from human activity recognition, smart devices, and advanced driver-assistance systems, among others.
Personal website: https://my591234.wixsite.com/islab
Sennott Square Building, Room 5317
Friday, January 20 at 2:00 p.m. to 3:30 p.m.