3D Human Activity Recognition by Indexing and Sequencing (RISq)

Human activity recognition has been a crucial area of research in computer vision over the past several years. Myriad applications of human activity recognition have been the driving force behind this research. The goal of human activity recognition is to automatically examine and classify activities in a 2D or 3D video sequence. A good deal of state-of-the art recognition algorithms has been researched and developed for this purpose, each of them handling the problem of human activity recognition in a unique way and achieving significant recognition rates. The advent of the kinect camera from Microsoft has been a breakthrough for the research on human activity recognition. The advantages of the 3D kinect camera over a 2D video cameras has enabled researchers to develop much more reliable recognition algorithms which are invariant to human position and orientation. These algorithms are capable of achieving correct recognition rates much higher than what was achieved in the past. In our work, we propose a novel approach for the problem of human activity recognition. We exploit the advantages of the kinect camera and the simplicity of our recognition algorithm: Recognition by Indexing and Sequencing (RISq). We capture the activity sequences using the kinect, and with the help of the software packages associated with the kinect we extract the skeletal information of the person. This information after processing acts as feature inputs for the RISq algorithm. The advantage of the kinect over conventional 2D cameras enables us to acquire this skeletal information with relative ease. With these features we train the RISq algorithm to recognize and classify an unknown activity sequence.