Compressed Domain Action Recognition for Healthcare and Assisted Living

Traditional action recognition methods aretime-consuming and need a high-performance hardwarefor required calculations. Nowadays in many popularapplications, compressed videos are available. Weproposed a method which uses available information incompressed domain and improves the performance ofextracting features from neural network by usingresiduals of frames instead of decoded and reconstructedframes. This work proposes a fast and efficientrecognition of activities. The proposed approach reducesthe computational cost of action recognition since thecompressed video information is explored. Lowcomplexity of the proposed method makes it properespecially for healthcare and assisted living purposes. Theexperimental results clearly in daily living datasetillustrate that the proposed low computationalcompressed domain approach provides acceptableperformance in terms of recognition accuracy.

Keywords:

action recognition , compressed domain , ADL , real time applications