本文主旨在發展整合手部膚色偵測技術,並藉由調整高斯膚色模型所求得的門檻值提高正確率,其最高平均正確可達98.64%。本研究的動機為針對智慧家庭系統中,以使用者的手部膚色為模型,並用較簡便的方式建立門禁系統,再經過調整與訓練,以同時達到高識別率與即時偵測能力。 而後與紋理特徵法結合,並使用Haar like特徵與Adaboost演算法並改變分類器,由傳統的串接式分類器改變為多類別分類器,能順利將三種手勢分類,而三種手勢最低誤差率分別0%、46.25%、17.5%。
This study is developing detection technology of hand skin color, and by adjust Gaussian skin color model obtained threshold exaltation correct rate, the highest average correct rate are 98.64%. This study the motivation is used in digital home-cared systems. And establish security system used method is more simple. Through adjustment and training. Concurrently achieve the higher recognition rate and real-time detection capability. Then with texture features integrate, and used haar like feature with adaboost algorithm and change classifiers, from traditional cascade classification transform multi-class classification, all right classification three sign language, the three sign language lowest error rate are 0%,46.25%,17.5%.