Human Activity Recognition using 3D CNN

Abstract

One of the most common biometric strategies is Human Activity Recognition, which is popularly known as HAR, which has gained immense popularity and significance over the last few years because of its applications. Human Activity Recognition deals with predicting the activity of a person based on the usual movements of humans such as sitting, standing, and so on. This paper presents an implementation of a Vision-Based Human Activity Recognition. The model is designed to predict the user’s physical activity with sufficient confidence and overcome the disadvantages of the sensor-based methods in terms of cost, need for fixed infrastructure, inaccurate results, and discomfort due to the physical contact with the person’s body. The proposed model uses convolution neural network architectures to meet the requirements. The 3D CNN model is used to implement the Vision-Based Human Activity Recognition. The UCF-50 dataset is used to train the model. The model will classify the images based on the activity that is being performed. Videos will be given as an input to the model and then the pre-processing will be done, which will be given to the model, and then the model will classify the activity. The vision-based human activity recognition system has numerous advantages in terms of accuracy, cost-effectiveness, and user-friendliness. It has a wide range of applications in elderly care, surveillance system, and anomalous behavior detection.