Abstract
Human gesture recognition allows for a more natural human machine interface eliminating expensive training for human's to get accustomed to the machines and avoid costly mistakes that follow till one becomes an experienced user. With advances in technology embedded devices with additional processing power and memory are becoming available. This is making our machines more capable and complex to operate, though the cost of human error is even higher. Hand gesture recognition offers a solution, but it still remains a very time and space complex problem when most non statistical methods are employed. Thus most embedded systems with limited space and processing power are unable to support hand gesture recognition. The paper introduces a statistical method which converts image contour to orientation based hash codes in-order to project it to a 3D-address space bounded by hamming distance. The main objectives are to reduce time, space complexity along with complete rotation invariance and online scalability. The implemented method proved to be 82.1% accurate against 1000 images comprising of 10 distinct static hand gesture sets.