A scientist from Samara University has developed a new method for recognizing human actions based on motion capture technology.
Unlike existing approaches that process huge arrays of data from sensors, the new technique transforms movements into compact trajectories formed from a sequence of human poses. This allows the computer to more accurately determine when and what action is performed, even if it is performed at different speeds or amplitudes.
Data on human movement, obtained using video or motion capture systems, is redundant. Therefore, we are trying to find a compact representation of this data that will allow the machine to accurately recognize human actions
The development has a wide range of applications. In medicine, it can be used to diagnose disorders of the musculoskeletal system, in security systems — to recognize suspicious actions, and in smart homes — to automate interaction with humans. The next stage of research involves recognizing movements from video and creating our own dataset for testing methods.
The novelty of the approach is important because it allows making action recognition technologies more accurate and versatile. This speeds up information processing, reduces the load on systems, and opens up opportunities for implementation in medicine, security, and household smart systems.
