ㅇ 제목 : Person Re-identification with Pose Priors and Camera Network Topology Inference in Large-scale Camera Networks

ㅇ 일시 : 2018년 5월 18일(금) 14:00 - 16:00

ㅇ 장소 : ETRI 융합기술연구생산센터 224호실

ㅇ 강사 : KAIST 윤국진 교수

ㅇ 요약 :

In this talk, I will introduce three person re-identification methods to overcome the limitations of the previous works. The main ideas of our methods are to infer contextual information from the scene (e.g., people poses, people walking speeds, spatio-temporal relationships between cameras) and to exploit the information for performing efficient person re-identification. First, we propose a novel framework for person re-identification by analyzing camera viewpoints and person poses called Pose-aware Multi-shot Matching (PaMM). It robustly estimates people poses and efficiently performs multi-shot matching based on the pose information. Next, we propose a unified framework which jointly solves both person re-identification and camera network topology inference problems with minimal prior knowledge about the environments. The camera network topology represents a spatio-temporal relationship between non-overlapping cameras. The proposed framework takes general multi-camera network environments into account and can be applied to online person re-identification in large-scale multi-camera networks. Lastly, we propose a novel distance-based camera network topology inference method for efficient person re-identification. We first calibrate each camera and estimate relative scales between cameras. Using the calibration results of multiple cameras, we estimate the speed of each person and infer the distance between cameras to generate distance-based camera network topology. The proposed distance-based topology can be applied adaptively to each person according to its speed and handle diverse transition time of people between non-overlapping cameras. To validate the proposed methods, we extensively compare with state-of-the-art person re-identification methods using various public datasets: 3DPes, PRID2011, iLIDS-VID, MARS, MCT and SLP. In our experiments, the proposed methods outperform the state-of-the-art methods in terms of both speed and accuracy.