Vision-based Activity Recognition: Present Status & Future Challenges

Abstract

Video, skeleton joint points are widely explored for human activity recognition (HAR). On the other hand, various sensors are engaged in human activity and behavior understanding. Vision-based human action or activity recognition approaches are based on RGB video sequences, depth maps, or skeleton data – taken from normal video cameras or depth cameras. On the other hand, sensor-based activity recognition methods are basically based on the data collected from wearable sensors having accelerometers, gyroscopes, and so on. There are numerous applications on HAR, however, healthcare, elderly support, and related applications become very important arenas with huge social and financial impacts. Due to the advent of various IoT sensors, it becomes more competitive as well as easier to explore different applications. The keynote will cover our works related to HAR approaches, highlighting healthcare perspectives and methods.