Development of models for classifying the movements of an anthropomorphic body from a video stream

Abstract

Objective. Today, capture is a chain for the implementation of medical rehabilitation systems, systems for measuring human physical activity and other medical applications. Their solutions often use hardware systems - sensors, which have a set of limitations and reduce the efficiency of access systems, increasing their cost. The following goal is required: Increasing the availability of application systems being developed, achieving steps without increasing the number of restrictions.Method. To achieve the goals given in the article, the following approach is used, based on processing a video stream from a camera that records the spectrum of visible radiation. During the research, a set of experimental data was collected.Result. As a result, a method for classifying video images of a visible phenomenon was developed, which differs from the use of existing models to detect key points of an anthropomorphic body in an image.Conclusion. This method avoids the use of special equipment and sensors (for example, the Kinect infrared camera) to implement application systems, increasing the availability of such systems and recording their special limitations.

Keywords

• human pose estimation
• movement classification
• fully-connected neural network