Novel Algorithm for Multi Hand Detection and Geometric Features Extraction and Recognition

Abstract

The recent trends for house appliance devices are moving towards the natural interaction that ensures free-cumbering interaction with these human-made devices, we have presented herein our novel approach for features extraction template that enables such devices to interact in better way especially when gesturing to vision-based devices such as home TV, these vision-based devices preferred over tis competitions since its no-frills communication, robot and video game interaction without any sensors or extra hardware just like human communicates with his same species, our algorithm extracts the hand gesture structure which are palm, wrist and fingers with their corresponding features like their locations, fingertips locations, finger bases locations, wrist location and their order from left finger to right finger regardless the hand orientation even upside down or any other angles, these features are important for latter approaches for vision-based algorithms and hand/palm/fingers tracking algorithms, we have classified the finger(s) according to their five classes that are thumb, index, middle, ring, and pinkie; these fingers have been classified using Gaussian likelihood function as a classifier regardless which hand is presented left or right and without any prior assumption of the pose of the hand, any hand and any orientation, after the finger's classification we have proceeded with hand gesture recognition by preserving one binary bit for every finger and gesture indexing done perfectly, we have considered all the outcomes of such finger's raising that are 32 combinations, our system can detect multi-hands that contained in a single image frame as well as recognition step, we have applied circular templates by using dynamic template matching with two different radiuses for each of fingers and palm premises respectively, we have achieved a perfect classification of multi-hands with their corresponding palm/fingers features for different samples as well as the recognition results and we have listed the recognition percentages for finger-wise and hand-wise as well as their processing time.