Sign Language (SL) allows hearing and speech impaired people to communicate each other. Nevertheless, SL is not understood by most of common people. The development of a real-time Mexican Sign Language (MSL) translator could benefit the speech impaired community. In this work we propose a method capable of recognize in real-time a basic list of Mexican Sign Language (MSL) signs of 20 meaningful words and translate them into speech and text. The signs were collected from a group of 35 MSL signers executed in front of a Microsoft Kinect Sensor. The hand gesture recognition system use the RGB-D camera to store depth, color and skeleton tracking information. We propose a method to obtain the representative hand trajectory pattern information. A Dynamic Time Warping (DTW) algorithm is used to interpret the hand gestures. Finally, we use K-Fold Cross Validation method for testing stages. Our results achieve a mean accuracy of 99.1% using N-best strategy (N=5) after find the best-match approaches of our templates data-set. And a mean accuracy of 98.57% from real-time testing.

Modelado 3D del lenguaje de señas mexicano para un sistema de voz-a-lenguaje de señas

In this work, the sensorless speed control of a brushless direct current motor utilizing a neural network is presented. This control is done using a two-layer neural network that uses the backpropagation algorithm for training. The values provided by a Proportional, Integral, and Derivative (PID) control to this type of motor are used to train the network. From this PID control, the velocity values and their corresponding signal control (u) are recovered for different values of load pairs. Five different values of load pairs were used to consider the entire working range of the motor to be controlled. After carrying out the training, it was observed that the proposed network could hold constant load pairs, as well as variables. Several tests were carried out at the simulation level, which showed that control based on neural networks is robust. Finally, it is worth mentioning that this control strategy can be realized without the need for a speed sensor.

In this work we used the Kinect® sensor in order to obtain tridimensional information to perform hand pose recognition. This recognition was used to implement a system that identifies all the hand poses of the Mexican Sign Language (MSL) alphabet. We used the fusion information that provides the IR and RGB cameras in order to determinate the finger's positions and assign a skeleton to the 3D data that belongs to the hands. We take into account the distances between a reference point and the phalanges as feature to distinguish among the symbols of the MSL. In order to perform hand pose recognition with the system, a three-layer neural network with backpropagation learning was implemented. The system was tested in real time with a user different from the one used to train the system, obtaining a recognition ratio of 90.27%.