Action recognition in videos

Human action recognition in videos is becoming more and more popular in applications such as intelligent surveillance, automatic video annotation and multimedia information retrieval. In this report, an action recognition algorithm based on supervoxel segmentation and bag-of-words representation will be introduced. The algorithms first segments the videos in supervoxels, and then extracts several types of visual features from the supervoxels. These extracted supervoxels features are then clustered to get a codebook to code bag-of-words representation of each video. Finally, the bag-of-words representation are trained with machine learning classifiers such as support vector machines with kernel methods such as linear kernel and chi-square kernel to classify human actions in new videos.Bachelor of Engineerin