Human Action Recognition using Computer Vision and Deep Learning Techniques

Abstract

Activity detection is a significant problem in smart video monitoring. The ability to recognize human movement in surveillance footage is a complicated computer vision task, however, owing to the size of the video as well as the CCTV camera's high resolution, doing so typically takes a long time. Therefore, it's necessary to lower the video clips resolution and identify the subjects' current activities. Using a variety of video processing methods and Convolutional neural networks associated with long-short term memory networks are examples of deep learning methods. We develop a model which uses the above networks for human action recognition on videos. Convolutional LSTM and Long Recurrent Convolutional Networks (LRCN) are the two models used for detecting human actions in videos and those models will utilize both the Convolutional Neural Network (CNN) and Long-Short Term Memory Network (LSTM). Both approaches can be used with Tensor Flow. The best model will then be taken into consideration for making predictions on Y ouTube videos. There are a lot of deep learning systems available right now, however, none of them work well when the video has a lot of details and it becomes challenging to identify the actual activity, Therefore, We select Convolutional Neural Networks (CNN) for picture data and Long-Short Term Memory (LSTM) networks for sequence information. ConvLSTM and LRCN have accuracy rates of 72.34% and 94.89%, respectively. video surveillance, recognizing human activity, patient monitoring, and human-robot interaction are some of the social effects of this research.