https://scholars.lib.ntu.edu.tw/handle/123456789/148644
標題: | 利用基因演算法建立人體姿態參數模型 Extraction of Parametric Human Posture Model Using Genetic Algorithm |
作者: | 徐仲安 Hsu, Chung-An |
關鍵字: | 姿勢評估;人體行為辨識;基因演算法;human behavior identification;posture determination;genetic algorithm | 公開日期: | 2004 | 摘要: | 辨識人體運動行為是現今電腦視覺中一項有趣的研究主題,已被廣泛地運用在虛擬實境、監控系統、人機介面、人體動作分析與多媒體壓縮等應用上。在含人體運動的連續畫面中,辨識人體運動行為主要是包含兩項步驟: 首張畫面的姿勢估測和使用連續的畫面做人體行為辨識。決定人體的姿勢參數需要搭配一個人體模型與目標影像進行逼近,現有文獻使用參數逼近的方法建構人體運動模型,卻會因為人體模型的參數設定過多,導致逼近的結果不佳。 為了能夠觀察出人體的姿態以及辨識出人體的運動行為,本篇論文中我們提出一套獲取三維人體模型關節參數的系統。做法是先建構出一個近似真實人體外型之三維人體參數模型,接下來分為兩個步驟。第一步、在固定攝影機下,從背景中擷取出人體的輪廓,在這個方法中,我們可以即時地更新背景模型資訊,以及得到運動中的人體輪廓。第二步、再利用基因演算法來將人體輪廓比對三維人體模型以產生最適當的關節參數。基因演算法擁有搜尋全域最佳解的特性,因而能夠解決大量參數最佳化的問題。基因演算法透過對於母代做選擇、交配以及突變等運算子來產生下一個子代。經過了適當的演化次數,經由適應函數的評估,保留下最好的個體,即可以得到適合的關節角度參數,將從影像中所得到的人體輪廓,轉換為一個具有關節角度參數的三維人體模型。實驗中將證明本方法可以成功地將數張影像中的人體輪廓擷取出來,並且可以將其轉換為具有關節角度參數的三維人體模型。 Recognition of human behavior has become an interesting research topic in computer vision, having a wide range of applications in virtual reality, surveillance systems, user interfaces, human motion analysis, etc. Recognition of human motion mainly involves two steps: posture determination of the initial frame and continuous motion identification of following frames. In order to obtain the human posture parameters of the target, a parameterized artificial human model is necessary to fit the image feature of the target. The existing methods for human motion modeling which required a large number of parameters of human model, it is difficult to obtain accurate estimation results. We present in this paper an approach to extract human parametric 3-D model for the purpose of estimating human posture. At first, a generic parameterized human model is developed. Then the task is done in two steps. In the first step, human silhouette is extracted from background under a fixed camera through a statistical method. By this method, we can reconstruct the background dynamically and obtain the moving silhouette. In the second step, genetic algorithm is used to match the silhouette of human body to a model in parametric shape space. The evolution ability of genetic algorithms can solve large parameter optimization problems. Genetic operations such as natural selection, crossover, and mutation are performed, and individuals in the next generation are generated. After a certain number of repetitions for these processes, the estimated parameter values are obtained from the individual with the best fitness. Experiments using human images show the promising results. |
URI: | http://ntur.lib.ntu.edu.tw//handle/246246/53253 | 其他識別: | en-US |
顯示於: | 電機工程學系 |
檔案 | 描述 | 大小 | 格式 | |
---|---|---|---|---|
ntu-93-R91921046-1.pdf | 23.31 kB | Adobe PDF | 檢視/開啟 |
在 IR 系統中的文件,除了特別指名其著作權條款之外,均受到著作權保護,並且保留所有的權利。