Hand gesture detection by genetic algorithm and multilayer perceptron
محورهای موضوعی :Fatemeh Mirzaei Asl 1 , Saman Shojae Chaeikar 2 , Saeid Yazdanpanah 3 , Arman Roohi 4
1 - Department of Computer Engineering Khorramabad Branch, Islamic Azad University
2 - Sydney International School of Technology and Commerce Sydney, Australia
3 - Department of Computer Engineering Khorramabad Branch, Islamic Azad University Khorramabad, Iran
4 - Sydney International School of Technology and Commerce Sydney, Australia
کلید واژه: hand pose detection, finger bent angle, genetic algorithm, MLP,
چکیده مقاله :
Recently, controlling robots through remote user hand motions has turned into an important research area in robotics. Robots normally employ a combination of image processing and learning techniques to detect hand poses, and metaheuristic techniques that follow the natural rules or behavioral models of living things can help to enhance their performance. This paper proposes a supervised learning method that applies genetic algorithm principles on a multilayer perceptron to improve the performance of the current MLP-based algorithms for hand pose detection. The method is trained and evaluated on a dataset consisting of data acquired from 300 users. RMSE benchmark reports an index value of 0.0424, and the experiments show a sensitivity of 93.42% and accuracy of 92.27% – 10.25% and 6.65% improvement compared to the MLP implementation.
Recently, controlling robots through remote user hand motions has turned into an important research area in robotics. Robots normally employ a combination of image processing and learning techniques to detect hand poses, and metaheuristic techniques that follow the natural rules or behavioral models of living things can help to enhance their performance. This paper proposes a supervised learning method that applies genetic algorithm principles on a multilayer perceptron to improve the performance of the current MLP-based algorithms for hand pose detection. The method is trained and evaluated on a dataset consisting of data acquired from 300 users. RMSE benchmark reports an index value of 0.0424, and the experiments show a sensitivity of 93.42% and accuracy of 92.27% – 10.25% and 6.65% improvement compared to the MLP implementation.
