Currently, human-computer interfaces have a number of useful applications for people. The use of electromyographic signals (EMG) has shown to be effective for human-computer interfaces. The classification of patterns based on EMG signals has been successfully applied in various tasks such as motion detection to control of video games. An alternative to increasing access to these applications is the use of low-cost hardware to sample the EMG signals considering a real-time response. This paper presents a methodology for recognizing patterns of EMG signals given by arm movements in real time. Our proposal is based on an artificial Neural Network, Multilayer Perceptron, where the EMG signals are processed by a set of signal processing techniques. The hardware used for obtaining the signal is based on Ag/AgCl connected to the EKG/EMG-Shield plate mounted on a Arduino One R3 card which is used to control a video game. The implemented application achieves an accuracy above 90% using less than 0.2 s for recognition of actions in time of testing. Our methodology is shown to predict different movements of the human arm reliably, at a low cost and in real time.