以顏色特徵為基礎的人物偵測

本文提出以顏色特徵為基礎之新的人物偵測方法。該方法採用正規化RGB色彩的表示方式，以降低系統對光線變化的敏感度及整合臉部及身軀偵測的策略；詳細步驟為利用臉部顏色的偵測找出最大膚色區塊及嘴唇顏色偵測決定正確臉部位置，並適當的設計身軀偵測，如身穿藍色衣服等，即可達成人物偵測；同時為去除雜訊，採用形態學的運算子，並利用連通標記法，以獲得完整的人物偵測。使用這些步驟，使得我們提出的方法，具有高效率而低複雜度的優點。經實驗證實所提出的方法非常有效率且實用。

關鍵字

顏色特徵；顏色分割；人物偵測

並列摘要

In this paper, we present a novel human detection method based on color features. Our approach has advantages of high efficiency and low computation load. In order to reduce the sensitivity of the proposed algorithm to illumination variations, a normalized RGB technique is used to represent a color image. An integration of face and body detections is employed in our approach. The detailed procedures are as follows: First, a broad area of skin color of a face can be detected by using a feature of the distinctive color to the background area; second, a lip can be detected by adopting its darker skin color in contrast to that of the face; third, the face is located according to its relative position to the lip; finally, the body can be detected easily by wearing appropriate color of clothes such as blue color. In addition, the morphological operations are applied to remove noise and label connection scheme is used to obtain complete human detection. Experimental results demonstrate that the proposed approach is highly efficient in computations and practical in applications.

並列關鍵字

color feature ； color segmentation ； human detection

主題瀏覽