Human Action Recognition Based on Depth Images from Microsoft Kinect

Recognition and Modeling of Atypical Children Behavior

“…The new sample is classified to cluster with the lowest distance and the appropriate code is assigned. In [12] the parameter k is set to 5 as the optimal value, therefore each action to be recognized comprises five stages. In [16] the k-means algorithm has been applied to histograms clustering problem in order to obtain HMM symbols (clustering centers).…”

“…Because of this convenience and broad perspectives presented in [7], [9], [12][13][14] it has been decided to use Kinect as data source. The system for autistic children proposed in this paper may be divided into subparts by their requirements.…”

“…In contrast to previously described system's subpart, the information about activity that is being performed is not given. The problem is similar to the issues depicted in [6], [8], [10][11][12], [14][15][16] where the solution is based on HMMs. HMMs' hidden states may be understood as the stages of modeled activities ( [8], [11], [15]) where each state's output is a discrete probability distribution [11].…”

“…The feature vector consists of 8 pairs of angles calculated for first-and second-degree joints in the way dependent on the parent group. In [12] for 7 right hand actions recognition two normalized vectors are used -each designated by two joints belonging either to a forearm or an upper forearm. Also [13], for the gesture recognition problem uses the skeleton structure included in Microsoft SDK.…”

“…There are also many studies motivated by game development and human computer interaction, e.g. [12,13]. Apart from research aimed at specified applications a lot of articles focus on action recognition just as the issue of methods: [14][15][16].…”