1 Contribution Human Body Motions Classification J. Havlik, J. Uhlir and Z. Horcik FEE CTU in Prague/Department of Circuit Theory, Technická 2, Prague 6, Czech Republic Abstract— This paper deals with video based parameteriza- tion and classification of human body motions. The main task of this work is to develop and verify the procedures for observing of muscle and brain activity. The developed procedures have no negative impact to brain activity (the tracking does not affect the measured EEG signals). The procedures required only standard hardware equipment accessible on neurological laboratories. The body motions are non-contact sensed using a pair of standard DV camcorders. This work includes the description of observing, discerning and parameterization procedures and the discussion of motion classification. The set of classifiers – hierarchical clustering algorithm, recursive clustering algorithm, k-means classifier, Bayes classifier and classifier based on discrimination functions – was developed and implemented. The analysis of the classifiers properties was accomplished in this work. The accuracy of classification was tested for selected classifiers. Keywords — motion analysis, motion classification, image processing I. INTRODUCTION This paper describes procedures for sensing, parameteri- zation and classification of human body motions. The main task of presented work is to develop and verify the procedures for observing of muscle activity. Presented work is a part of study of correlation between a human brain and muscle activity. The brain activity is represented by the electro–encepha- lograph (EEG) signals for this research. The EEG signals are very complex signals reflecting not only intentional motions, but also all vital functions, artifacts from eye motions etc. Due to this complexity it is appropriate to use a small peripheral muscle capable of independent motions to minimize undesirable effects of its neighbourhood. Relatively simple brain stimulus could be expected for this movement. Due to these reasons the free three–dimensional motions of thumb have been chosen. The muscle activity is represented by the parameters of the thumb trajectory. The goal of current research is to assign typical changes of EEG signals to the type of thumb motion. The parameterization and the classification of thumb motions are the aim of presented work. II. EXPERIMENT A sensed person seats in straight seat. The arm with observed thumb is supported by a rest. The thumb moves between 3 positions – stationary states. Each move is triggered by the synchronization pulse. The period of pulses is 6 ± 1 seconds. About 20 % of period is a motion. The rest of period is a stay on the position. The recording of motions has to fulfill two necessary conditions: 1. a non–contact sensing and 2. a possibility to synchronize the recorded movements and EEG signals. The sensing has to be non–contact, because any contact between moves thumb and any part of sensor could damage the EEG signals due to a physiological feedback. The recorded moves and the EEG signals are processed separately. The both records have to be synchronized in order to study the correlations. According to these conditions the thumb motions are sensed using a pair of standard DV camcorders. The motions are triggered with LED generated optical pulses, the leading edges of synchronization signal are recorded as a polygraphical signal parallel to EEG signals. III. PARAMETERIZATION The sensed thumb is marked by the special mark, the black and white concentric circles. The outputs of recording are two video sequences in PAL standard stored on a tape. The main task of the parameterization is to find the parametric description of thumb motion. The parameterization process could be separated to the three parts: 1. the preprocessing of input video-sequences, The input video-sequences are transformed to B/W images with white background and black regions in places of special marks.