應用主動式深度攝影機於人體姿勢自動辨識系統研究

A Study of Automatic Human-Pose Recognition System using Active Depth Cameras

陳昱辰(Yu-Chen Chen)

指導教授 ： 張元翔

中原大學/工學院/資訊工程研究所/碩士(2012年)

https://doi.org/10.6840/cycu201200674

摘要

由於主動式深度攝影機 (如：微軟Kinect攝影機) 的技術突破，可以同時提供色彩影像、深度資訊及骨架特徵，人體物件的分割比傳統2D電腦視覺技術更為精確。本研究的目的便是採用主動式深度攝影機，藉以開發一套可以辨識五種特定人體姿勢 (即無姿勢、生氣、高興、懊惱、無奈) 的自動人體姿勢辨識系統。系統主要步驟包含：(a)深度影像前處理；(b)骨架特徵擷取；(c)定格姿勢偵測；(d)姿勢估測及辨識。我們蒐集一組資料庫作為系統訓練及測試之依據。初步的研究結果顯示，本系統可以達到88%的整體辨識率。總結而言，本系統在人體姿勢的自動辨識具有潛力，因此可以用來增進人機互動。

關鍵字

電腦視覺 ； 主動式深度攝影機 ； 視訊處理 ； 人體姿勢辨識

並列摘要

Because of the technical breakthrough of active depth cameras (e.g., Microsoft Kinect cameras) that can simultaneously offer color image, depth information, and skeletal features, segmentation of human objects has been relatively accurate than in conventional 2D computer vision techniques. Using such an active depth camera, the objective of this study is aimed to develop an Automatic Human-pose Recognition System for five specific human-poses (i.e., no pose, angry, happy, annoyed, and helpless). The man processes include: (a) Depth Image Preprocessing; (b) Skeletal Feature Extraction; (c) Freeze Pose Detection; (d) Pose Estimation and Recognition. A database was collected for the system training and testing. Our preliminary results indicate that our system could achieve an overall detection rate of 88%. In conclusion, our system could be potentially used for automatic recognition of human-poses, thus leading to enhance human-computer interaction.

並列關鍵字

active depth camera ； video processing. ； human-pose recognition ； computer vision

參考文獻

[4] 朱陸中，多視角三維人體姿態追蹤–利用柔性關節規範之迭代最近點演算法，國立台灣大學資訊工程研究所，碩士論文，民國97年。

[6] J. O. Rourke and N. I. Badler, “Model-based image analysis of human motion using constraint propagation,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 2, no.6, pp.522-536, 1980.

[7] D. Hogg. “Model-based vision: A program to see a walking person,” lmage and Vision Computing, vol.1, no.1, pp.5-20, 1983.

[9] L. Kakadiaris and D. Metaxas, “Model-based estimation of 3D human motion,” IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 22, no.12, pp.1453-1459, 2000.

[10] W. Qu, and D. Schonfeld, “Real-Time Decentralized Articulated Motion Analysis and Object Tracking From Videos,” IEEE Transactions on Image Processing, vol. 16, no. 8, pp.2129-2138, 2007.