Human Activity Identification in Smart Daily Environments

Research in human activity recognition (HAR) benefits many applications such as intelligent surveillance systems to track humans’ abnormal activities. It could also be applied to robots to understand human activity, which improves smart home efficiency and usability. This chapter aims to accurately recognize different sports types in the Sports Video in the Wild (SVW) dataset employing transfer learning. The dataset consists of noisy and similar classes shot in daily environments, not in controlled laboratory environments. Heretofore, different methods have been used and developed for this purpose. Transfer learning is the process of using pre-trained neural networks. The experimental results on different splits of the dataset, size, and pre-trained models show that accuracy of 80.7% is achievable. In another experiment, we have used the famous UCF101 dataset which is collected from YouTube and trained a convolutional neural network (CNN) with Batch Normalization (BN). The achieved accuracy for the test dataset is around 91.2%. One application of the proposed system is to integrate it with a smart home platform to identify sports activities of individuals and track their progress.