Dynamic Recognition and Classification of Trajectories in SLRecon Adopted Artificial Intelligence in Kinect

Abstract

We have proposed “SLRecon” a digital representation of the exoskeleton by Kinect software to analyze the movement of the hands and thus identifies the trajectories taken by the signs for further processing. Subsequently, the trajectories were considered for phases such as training, validation and testing of a neural network-based artificial intelligence algorithm. The network responsible for recognizing and classifying 5 important signs determined by an expert. The neural network is a multilayer perceptron that was trained using the backpropagation method. The training phase was performed with 6 subjects and additionally tested with 9 subjects. We also discussed the results from the simulation phase, which confirmed that the system achieved 99.6% efficiency in detection and classification, while it achieved 98.7% accuracy in the field test. Finally, we compared and validated our results with other methods.