Fast, effective and robust person re-identification for large-scale real applications

2017 Fiscal Year Final Research Report

• Project/Area Number: 15K16024
• Research Category: Grant-in-Aid for Young Scientists (B)
• Allocation Type: Multi-year Fund
• Research Field: Perceptual information processing
• Research Institution: Nara Institute of Science and Technology
• Principal Investigator: Wu Yang 奈良先端科学技術大学院大学, 研究推進機構, 特任助教 (30750559)
• Project Period (FY): 2015-04-01 – 2018-03-31
• Keywords: Person Re-identification / Pattern Recognition / Computer Vision / Machine Learning / Deep Learning / Transfer Learning / Active Learning

Outline of Final Research Achievements

We investigated solutions for making person re-identification models more effective, more scalable, faster and more robust, toward being applicable to real world intelligent video surveillance scenarios. For enhancing the effectiveness, we have explored dictionary learning, metric learning, locality, and end-to-end deep learning. For efficiency and scalability, we proposed a hierarchical interclass structure learning model which allows fast search for classification even when handling large amounts of classes. For robustness, we designed one transfer learning model and two active learning models which allows across-dataset model transferring and training with minimum supervision. 10 papers have been published and all the proposed models have been evaluated with extensive experiments and comparisons.

Free Research Field

コンピュータビジョン