Deep Learning Based Human Activity Recognition With Improved Accuracy

Perceiving human exercises from video clips or still pictures is a provoking mission because of issues like, changes in scale, perspective, lighting, and appearance of source images. Human action acknowledgment is a difficult time series order task. It includes anticipating the action of an individual in light of image sensor information and generally requires profound area mastery and techniques of image processing to accurately extract meaningful feature data from the crude information to fit an artificial intelligence model. Currently available models are exceptionally tedious and lack accuracy of classification result. So there is a need to plan a Human action acknowledgment model which can be accurate and can be utilized efficiently in present world applications. This model will not just be practical yet in addition will be a utility-based model that can be utilized in an enormous number of applications such as observing and caring home alone elderly people or monitoring any unattended patient in a hospital. In this proposed model, source video dataset is wisely prepared for a meaningful and concise feature extraction by techniques like optical flow and 2D spatial temporal feature extraction. Then, these features are fed to the model for training by a VGG-19 Algorithm to effectively increase the accuracy of the Human Activity Recognition model compared to the existing system.