基於雙核心平台的嵌入式步態辨識系統 ; Embedded Gait Recognition System Design Based on Dual-Core Platform

NCUIR > College of Electrical Engineering & Computer Science > Graduate Institute of Computer Science and Information Engineering > Electronic Thesis & Dissertation > Item 987654321/44707

Please use this identifier to cite or link to this item: http://ir.lib.ncu.edu.tw/handle/987654321/44707

Title:	基於雙核心平台的嵌入式步態辨識系統;Embedded Gait Recognition System Design Based on Dual-Core Platform
Authors:	蔡昌成;Chang-Cheng Tsai
Contributors:	資訊工程研究所
Keywords:	步態辨識;機率神經網路;粒子群最佳化;嵌入式系統;異質雙核心處理器;probabilistic neural network;particle swarm optimization;embedded system;heterogeneous dual-core processor;gait recognition
Date:	2010-07-27
Issue Date:	2010-12-09 13:53:21 (UTC+8)
Publisher:	國立中央大學
Abstract:	由於人們對於人身財產及居家安全的要求，視訊監控的需求越來越多。但是傳統的視訊監視只能錄下影片，無法針對影像資訊進行即時分析和決策。因此無需人員監看，能夠自動分析影像並提供決策資訊的智慧型視訊監控系統越來越受到大家的關注。本論文因此提出了一個以人類行走步態做為生物特徵的身份辨識系統，針對視訊監控應用，可提供即時影像分析並辨識目標物的身份。本研究的步態辨識流程分為步態影像切割、步態特徵抽取、以及機率神經網路分類器的步態辨識。步態影像切割需對連續變動的場景建立背景模型，並進行影像相減以得到前景物。特徵抽取的方法則是對步態影像進行水平投影與垂直投影，並將連續的投影結果進行時間軸的離散傅立葉轉換，得到步態特徵向量。最後我們使用機率神經網路做為步態辨識的分類器。此外，我們還使用了粒子群最佳化來對機率神經網路分類器的平滑參數進行最佳化，以得到最好的辨識效果。實驗結果顯示，我們所提出的步態辨識方法具有很高的辨識性能。最後我們將整個步態辨識的演算法移植到具有ARM處理器及DSP處理器的雙核心嵌入式系統上，分析每個演算法模組的計算複雜度，最後將所有模組切割分配在雙核心系統上，以得到最佳的嵌入式系統效能表現。Because people attach great importance to property and home security, the demand for video surveillance increases gradually. However, traditional video surveillance can only record videos, lacks the ability of instant analysis and decision for videos. Accordingly, people pay more and more attention to the development of intelligent video surveillance system. As a result, we propose a gait-based person identification system which can offer real-time video analysis and recognize person identification. The procedures of our gait recognition system are gait image segmentation, gait feature extraction and probabilistic neural network classifier(PNN classifier). In gait image segmentation, we build a background model to get the foreground object by image subtraction. In gait feature extraction, we compute the vertical and horizontal projection of gait images, and perform Discrete Fourier Transform (DFT) to the projection result. After DFT, we use the result as our gait feature vector. Finally, we use PNN classifier to perform gait recognition. Additionally, we optimize the smoothing parameters of PNN by particle swarm optimization (PSO) in order to get the best recognized performance. The experimental results show that our proposed method reaches high recognized performance. We then implement our gait recognition system on the dual-core embedded system OMAP3530, which contains an ARM processor and a digital signal processor. We analyze the complexity of every function in our method, and let the function with higher complexity perform on DSP to reach excellent system performance.
Appears in Collections:	[Graduate Institute of Computer Science and Information Engineering] Electronic Thesis & Dissertation

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	603	View/Open

社群 sharing

Loading...