Human motion analysis is an emerging research field for the video-based applications capable of acquiring and recognizing human motions or actions. The automaticity of such a system with these capabilities has vital importance in real-life scenarios. With the increasing number of applications, the demand for a human motion acquisition system is gaining importance day-by-day. We develop such kind of acquisition system based on body-parts modeling strategy. The system is able to acquire the motion by positioning body joints and interpreting those joints by the inter-parts inclination. Besides the development of the acquisition system, there is increasing need for a reliable human motion recognition system in recent years. There are a number of researches on motion recognition is performed in last two decades. At the same time, an enormous amount of bulk motion datasets are becoming available. Therefore, it becomes an indispensable task to develop a motion database that can deal with large variability of motions efficiently. We have developed such a system based on the structured motion database concept. In order to gain a perspective on this issue, we have analyzed various aspects of the motion database with a view to establishing a standard recognition scheme. The conventional structured database is subjected to improvement by considering three aspects: directional organization, nearest neighbor searching problem resolution, and prior direction estimation. In order to investigate and analyze comprehensively the effect of those aspects on motion recognition, we have adopted two forms of motion representation, eigenspace-based motion compression, and B-Tree structured database. Moreover, we have also analyzed the two important constraints in motion recognition: missing information and clutter outdoor motions. Two separate systems based on these constraints are also developed that shows the suitable adoption of the constraints. However, several people occupy a scene in practical cases. We have proposed a detection-tracking-recognition integrated action recognition system to deal with multiple people case. The system shows decent performance in outdoor scenarios. The experimental results empirically illustrate the suitability and compatibility of various factors of the motion recognition.
展开▼
