Implementation Of A DSP-based Gesture Interaction System

Gesture-based interaction provides a natural and harmonious method of human-computer interaction, and which has become one of the most popular research topics in the field of multimodal interaction. Compared to other well-researched gesture recognition technologies, gesture interaction based on SEMG and ACC signals has more advantages, such as low cost and wearable property. SEMG signals obtained by electrodes from skin surface contain vivid information of subtle gesture execution, while acceleration signals describe the large scale of gesture trajectory. Gesture recognition technology based on the fusion of the SEMG and ACC signals is more suitable for prosthetic control, sign language recognition, human-computer interaction, and gesture-based control of entertainment electronic products, etc.Based on the research progress obtained in the past years,, this thesis proposes a scheme of embedded platform, which realized the gesture recognition technology based on the fusion of SEMG and ACC signals, and designs a set of portable real-time gesture recognition system implemented on fixed-point DSP. In the realized system, the acquired data is transceived by the wireless 2.4G the gesture recognition algorithm is programmed on the fixed-point DSP. and the recognition results are translated into control commands, which are used to real-time control of household appliances. The work of this thesis lays a foundation for the technology promotion of human-computer interaction, and it's also a meaningful attempt to push forward the technology realization for the common use.The main work of the thesis can be summarized as follows:(1) Design of a set of real-time gesture interaction platform based on fixed-point DSP. Its hardware consists of 5 channel SEMG sensors, 1 channel 3-D accelerometer, a wireless receiver, a real-time signal processor based on DSP, and an interactive application module, etc. Eight kinds of common gestures are recognized through the configuration software, and the recognition results can be displayed on LCD or translated into control commands for external device.(2) Transplantation of the real-time gesture recognition algorithm on DSP platform. The algorithm includes segmentation, feature extraction and classifier. The common moving window method is selectedfor active segment detection. Considering the real-time property and high accuracy, the time-domain MAV (Mean Absolute Value, MAV) and the frequency-domain AR (Auto Regressive, AR) coefficients are chosen as the features of SEMG signals. As for ACC signals, the features are obtained by the down-sampling of the original signals. Meanwhile, the easier and more stable LDC (Linear Discriminant Classifier) has been excuted for recognition task.(3) Implementation of two kinds of interaction solutions based on gestures. One is the real-time remote application of the air conditioning based on the infrared encoding, the other is the image interaction application on the TFT LCD for the illustration of the recognition results.