Efficient human action recognition Algorithms using sparse Representation

Abstract

Human action recognition refers to the technology involved in understanding human behavior using attributes derived from various sensors. Various schemes involved in human action recognition can be broadly categorized as, vision-based and sensor-based schemes. Vision-based schemes refer to the systems that utilize data from cameras mounted at predefined locations whilst sensor-based schemes make use of data from sensors such as accelerometer, gyroscope, etc., that are attached to the bodies of individuals. The omnipresence and fall in the production cost of wearable sensors have resulted in the emergence of remarkable research contributions in the discipline of human action recognition using wearable sensors. Owing to these benefits, action recognition systems based on wearable devices have paved a path for a host of health-care applications like elderly care, assistive living, fitness tracking and so on. The achievement of high levels of accuracy is essential especially in medical applications so that, they can be reliably implemented in real-time scenarios. In addition, the design of algorithms with minimum recognition time is vital. Action recognition frameworks based on sparse representation has been proposed in this research with the objective to produce very good recognition results along with minimal recognition time. newlineAn approach for recognizing human actions using data from a single tri-axial accelerometer based on compressive classification is presented. A novel chaotic map is proposed for reducing the dimension of accelerometer data. Using the low-dimensional data, time and frequency-domain features are extracted. newline