A Robust Feature Based Approach for Bare-Hand Detection under Practical Non-Ideal Conditions

摘要

Due to easy detection of contact based devices, color-markers or 3D cameras, state-of-art gesture models mostly employs them to make the gesture. Yet, their lack of user-friendliness limits their use in practical scenarios. Use of barehand to make a gesture, however, comes naturally to the user. However, computer-vision based bare-hand detection is a challenging task, affected by multiple environmental factors. Feature based object detection techniques are easy and robust solution to detection under various non-ideal conditions. In this study, an extensive comprehensive study is carried between two purely spectral (color) features and fourteen color-texture features. Models are developed and compared for different image sizes. The Classification models are developed using Naive Bayes classifier (Probabilistic view), Euclidean distance and Chebyshev distance models (Proximity view) and Real AdaBoost classifier. Experimental results showed that only 2 out of the 16 proposed features has performance less than 90% for hand detection under noisy conditions.