Comprehensive survey of human-activity detection and recognition with time-series model

Human activity recognition (HAR) - the exciting time series based classification task has been categorize into vision, sensor and hybrid approaches based on the input modality. The availability of these modalities to capture data for HAR is a motivation factor in building AI / ML enabled applications in emerging areas such as health care, surveillance, etc. Based on the input data captured, HAR helps in identifying the specific movement of an individual. It also helps in inferring the current behavior and goals of the human body depending upon the environment, through a series of observations. A generic process flow of HAR involves data acquisition followed by pre-processing; feature extraction, feature selection and time series based classification process. The pros, and cons of the HAR approaches were analyzed along with a detailed report on the datasets being used which provides a comprehensive review on the usage of appropriate ML algorithms and analyzes the risks in the existing HAR models for the future scopes in this field.