单 位：计算机学院

报告题目:Where am I– Visual place Recognition with Deep Convolutional Neural Networks

报 告 人: Professor Hong Zhang

报告时间:2017年2月21日上午10:00

报告地点：大学城广东工业大学工学一号馆216室

个人简介:

Dr. Hong Zhang received his Ph.D. degree from Purdue University in 1986 in Electrical Engineering, with a thesis on robot manipulation and force control. Upon completing post-doctoral training at the University of Pennsylvania, he joined the Department of Computing Science, University of Alberta, Canada in 1988 where he is currently a tenured Full Professor. Dr. Zhang’s research interests span robotics, computer vision, and image processing, and his current research focuses on visual robot navigation in its indoor and outdoor applications. Dr. Zhang holds an NSERC Industrial Research Chair in Intelligent Sensing, and is a member of the NSERC Canadian Strategic Network on Field Robotics. He is an associate editor of IEEE Transactions on Cybernetics, and the General Chair of 2017 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). He is a Fellow of the IEEE, and a Fellow of the Canadian Academy of Engineering. In China, Dr. Zhang holds an adjunct appointment with the Guangdong University of Technology under GDUT’s 100-Scholar Program.

Email:hzhang@ualberta.ca, hzhang@gdut.edu.cn

URL: http://www.cs.ualberta.ca/~zhang

报告摘要:

Visual place recognition (VPR) answers the question of whether the current view - of a robot or a mobile device - comes from a place or location that has been visited in the past. The ability to recognize a place visually is a crucial component algorithm in solving many problems in robotics and content-based image retrieval, among others. VPR is challenging when the current camera view has changed significantly from that in previous visits to the same place, due to variation in the camera location, lighting and weather conditions, etc. In this talk, I will describe our recent research that addresses VPR by exploiting the remarkable performance of deep convolutional neural networks (ConvNets). ConvNets have recently been shown to outperform by a significant margin other solutions that use traditional techniques in object detection and recognition, and condition-invariance of ConvNets is key to the success. Building on this success, our work further shows how to use ConvNet to solve VPR accurately. In addition, we employ efficient data structures in order to implement our VPR solutions efficiently. Finally, I will highlight how to integrate VPR in a robot navigation system for effective robot localization.

欢迎广大师生参加！