1. Introduction

Applications in various fields such as video games [4], dance education [18], [19], sports [3], medicine (mainly rehabilitation) [11], [21], self-driving cars and robotics [5], [6], [10], and security [1] require an improved understanding and representation of human locomotion.

Various types of human activities are categorized in: gestures, actions, interactions, and group activities [1]. Gestures are elementary movements of a body part, for example “waiving hand”. Complex activities composed from multiple gestures organized temporally are actions. Some examples of human actions are walking, running, jumping. Interaction is a human activity that involves two or more persons and/or objects. Group activities are activities that involve a group of persons and/or objects. The movements of body parts at joints are mainly rotational and take place in three reference planes: sagittal, frontal and transversal. These movements are flexion and extension (frontal axis), abduction and adduction (sagittal axis), medial and lateral (transverse axis).

Walking is a human action in which one foot touchdown is followed by the touchdown of the other foot in a continuous pattern [3]. One of the differences between walking and running is that in the running cycle for a period of time both feet are in the air [7]. Jogging is a form of running at a slower pace [15].

The type of human locomotion may be indicative for the type of activity, intentions, attitudes, circumstances perceived – e.g., dangers, and mood of the person. Consequently, the fast detection of the type and specifics of the locomotion may indicate dangerous situations, among others.

The analysis of the human movement is qualitative and quantitative. While the qualitative analysis of human movement is a non-numerically analysis and describes human actions as patterns, the quantitative analysis requires a large amount of measurements of the parameters which describe the human action and a computer to perform numerical calculation [3].

Human locomotion represents a complex kinematics, with all the body parts participating, involving a large degree of freedom. In [17], [18] and [19] the authors have proposed semantics of the human movement. A semantic representation of human motion in combination with a motion recognition algorithm for digital human simulation is presented in [16].

The combined use of fuzzy logic and complex dynamics has been advocated by several groups, for example [8], [9], [22], including in representing and classification of human movements [9], [21].

This article follows the track established by the papers [12], [13], [23]. Compared to those preliminary papers and to the literature, this article brings the following contributions: a complete statistical characterization of the movement, with the related database, including all angles between the relevant body parts, the derivation of the membership functions based on the statistics, the automatic clustering of the movement sequences in view of objective determination of the membership functions related to a more detailed analysis of the dynamics, and detection of abnormal movements applications, with relevance in the medical and security fields.

Throughout the paper, the main parameters of the human locomotion, as established in the literature, for example in [24], are taken into account. This study follows the same path by analyzing basic human locomotion and proposing a fuzzy ontology of it, based on video analysis.

The paper organization is as follows. Section 2 presents the video database and the method used to determine the parameters of the locomotion. In Section 3 a K-means algorithm was applied on extracted parameters. In the fourth section are presented the experimental results and the last section concludes the paper.

REFERENCES

1. AGGARWAL, J. K., M. S. RYOO, Human Activity Analysis: A Review, ACM Computing Surveys (CSUR), vol. 43(3), April 2011.
2. ARTHUR, D., S. VASSILVITSKII, K-means++: The Advantages of Careful Seeding, SODA ‘07: Proceedings of the Eighteenth Annual ACM-SIAM Symposium on Discrete Algorithms, 2007, pp. 1027-1035.