Fine-Grained Classification of Person Attributes with Surveillance Cameras

2017 Fiscal Year Final Research Report

• Project/Area Number: 16K12460
• Research Category: Grant-in-Aid for Challenging Exploratory Research
• Allocation Type: Multi-year Fund
• Research Field: Perceptual information processing
• Research Institution: Nagoya University
• Principal Investigator: Kato Jien 名古屋大学, 情報科学研究科, 准教授 (70251882)
• Co-Investigator(Kenkyū-buntansha): ワン ユ 名古屋大学, 国際開発研究科, 助教 (60724169)
• Project Period (FY): 2016-04-01 – 2018-03-31
• Keywords: 歩行者属性の詳細認識 / 歩行者認識 / 部分と全体を着目したCNN融合 / 超解像処理 / グラフィックモデル / 姿勢推定

Outline of Final Research Achievements

The objective of this research is to recognize fine-grained categories of person attributes from enormous surveillance videos. To achieve this goal, we focus on 4 kinds of pedestrian attributes including 14 categories. We developed five methods to enhance the accuracy of fine-grained classification, including (1) super-resolution based image processing, which helps to recover the image details; (2) patch dividing based feature extraction, which extracts features in a way that preserves the spatial layout of inputs; (3) fusing multiple CNN models, which acquires more detailed features; (4) pose-wise classifier sharing, which learns robust classifiers and makes robust predictions; and (5) graphical model based inference, which utilizes the interdependence between different subcategories to update raw estimations to better ones. We conducted experiments on a pedestrian data set, and confirmed superior performance of our approach based on these methods over the state-of-the-art.

Free Research Field

画像・映像の内容理解