New Approach to Recognizing Human Movements

A scientist from Samara University proposed a new approach to recognizing human actions, based on the motion capture technology. The results are published in The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences.

Recognizing the actions allows machines to interpret certain movements and respond to them. As explained at Samara University, the technology can be used in many areas of human activity, such as medicine, public security, games and entertainment.

Human-movement data can be obtained, in particular, by using motion capture. This technology works as follows: sensors to record every step, turn or tilt are attached to the human body. The data collected is then processed by specialized software. Results of the processing can be used for creating digital animations, as well as recognizing actions or studying a particular person’s movement parameters.

One of the problems with the current approaches to recognizing the actions is that sensors generate huge amounts of data. Besides, various peoples’ movements of the same action may differ markedly in speed or in amplitude. In order for the computer to understand that it deals with “walking”, “running” or “jumping”, effective ways of processing this data are needed.

“Human-movement data obtained by using video or motion capture systems is redundant. Therefore, we try to find a compact representation of this data, which will allow the machine to accurately recognize human actions, that is, to determine at what point in time and what action is performed by the person”, said Evgeny Myasnikov, the author of the study, Associate Professor at Samara University’s Department of Geoinformatics and Information Security.

The scientist proposed a new approach to recognizing human actions, based on the transformation of the initial motion-capture data into a sequence of trajectory points in the “compressed” space. It is formed from various poses taken by a person. That is, any action can be represented as a trajectory and a movement can be recognized by comparing such trajectories. The researcher used mathematical methods that make it possible to properly interpret movements even at different performance rates.

According to the author, the developed approach has a wide application capability. In medicine, it can help diagnose disorders of the musculoskeletal system. In the field of security - to identify suspicious activities and prevent crimes. Moreover, the technology may become part of the smart-home system.

“Compared with the alternative approach, our approach, on average, more often correctly determines, which action is performed by the person. Herewith, we used a well-known data set made freely available by other researchers. If it is about the project as a whole, we still have to prove the advantage of the approach being developed, by using other data and a wider range of tasks”, explained the expert.

The specialist added that the next stage of the study involves solving the problem of movement recognition based on video data.

“Besides, we want to create our own data set that will help researchers develop and evaluate the efficiency of their own approaches to recognizing human actions. We have the necessary equipment for this”, concluded Evgeny Myasnikov.

The conducted research is part of the project “Methods of forming characteristic descriptions of human actions”, supported by the Russian Science Foundation within the framework of the competition for conducting fundamental research by small scientific groups.

Source: ria.ru