Published Paper Details:

In compared to its analogue predecessors, the digital IP video system offers a wide range of features and capabilities, but it still faces significant difficulties in the creation and implementation of intelligent and efficient systems. This initiative aims to satisfy the rising need for automated analysis and comprehension of human behaviour. The presentation of a real-time human activity identification system built on global feature extractions. The suggested method calculates an image's characteristics from video sequences. The goal is to combine local and global properties to create a powerful real-time recognition system. We utilise the UT Dataset, which includes the following actions: embracing, handshakes, kicking, punching, and pushing. Convolution neural networks (CNN) are used to instantly identify various human actions. Similar videos based on action recognized in input videos are retrieved from the dataset as output.