Development of gesture database for an adaptive gesture recognition system

Abstract

Human gestural motion is one of the areas in studying human behaviour regardless the physical capability and intellectuality of each individual. In this research, the focus is to investigate human physical characteristics which contribute to the performance of gestural motions. Every person has different body structure and physical distinctive that can be determined by calculating the person’s body mass index (BMI) and measuring the size represented by the weight an geometrical gestures. The geometrical gesture databases are developed based on human body characteristic features. These gesture databases are utilized to recognize and identify an unknown gesture by gathering some information of human features for further analysis. A motion capture system was used to capture gestural motions. Three dimensional data obtained from motion capture system are analysed, classified and stored in the gesture database. The resampling algorithm is developed to diminish the excessive movement information which to be used in the represented form. Principal Component Analysis (PCA) is used to reduce dimension of data and classify the gesture data. PCA classifies three groups of people based on gestural motions of subjects. For further clarification, data inside the group database were tested for similarity and dissimilarity measured using Jaccard Similarity Measure; the result of total average is 90.8% dissimilarity of all five geometrical gestures between group #1, group #2 and group #3 for all the three axes: X-axis, Y-axis and Z-axis. Consequently, adaptive gesture recognition is introduced to select the suitable database especially for identifying unknown gestures inserted into the system. The result of recognition shows that recognition of individual database is 86.5%, group database 83.7% and the lowest is recognition of universal database which is 82.8%. The experimental result shows that the group database is preferable for an adaptive gesture recognition system.