Human action analysis using K-NN classifier

Human activity recognition is an important research area of computer vision. There is an urgent mechanism to automatically detect and retrieve semantic events in videos based on video contents. Low-level video sequence content is translated into high-level video sequence content is a interesting research topic in recent years. Its applications include automated video surveillance schemes, intensive care systems, airports, analysis of the physical condition of people and a variety of systems which include human-computer interfaces. Automatic recognition of high level activities which refers combination of multiple simple human actions. By recognizing high level human action from video sequence helps to construct many important applications. Human activity can be analyzed in many ways. In this paper, wave pattern is used to identify different human activities. This pattern is generated by using texture features (energy) for the extracted foreground objects. Various kinds of human actions are analyzed and classified. The system consists of following stages: object detection, feature parameter extraction, pattern generation and classification. Gaussian mixture model is used for extracting foreground object from background image. Pre-processing is done in order to reduce the noise content from each and every frame. Then wave pattern is generated by using energy function. K-Nearest Neighboring is used to classify the different human activities. From this wave pattern presence of objects, direction of moving object, no of objects presented in foreground frame are observed. By using this information human activities in each frame are classified using K-NN. Two different types of dataset namely KTH dataset and WEIZMANN dataset is used. The accuracy rate of the system is 93.33%.