Compressed Domain Action Recognition for Healthcare and Assisted Living

Traditional action recognition methods are time-consuming and need a high-performance hardware for required calculations. Nowadays in many popular applications, compressed videos are available. We proposed a method which uses available information in compressed domain and improves the performance of extracting features from neural network by using residuals of frames instead of decoded and reconstructed frames. This work proposes a fast and efficient recognition of activities. The proposed approach reduces the computational cost of action recognition since the compressed video information is explored. Low complexity of the proposed method makes it proper especially for healthcare and assisted living purposes. The experimental results clearly in daily living dataset illustrate that the proposed low computational compressed domain approach provides acceptable performance in terms of recognition accuracy. Keywords-action recognition; compressed domain; ADL ;real time applications. I. INTRODUCTION As the aging population increases in modern communities, the need for remotely protecting elderly and patient at the home or other places has recently increased. Generally, the aim of healthcare and assisted living is improving the quality of life and providing healthy living of older or impaired people by using information technologies and machine vision. Technology can support people affected by various physical or mental disabilities such as chronic disease or elders who have Alzheimer. In the last decade, artificial intelligence has emerged as a powerful tool and can be useful in many aspects of life, such as healthcare and assisted living. Smart technologies can be utilized as a means to improve the quality of care and wellbeing of dependent people [1].