3D body shape clustering based on PSO by multi-fitness function

Abstract

Human body shape plays an important role in many industries, especially in Healthcare. We have seen changes in the body size and shape of the Thai population from the data collected through national sizing surveys over two decades. These changes can also indicate the trend of national health status. In this paper, a body shape clustering algorithm based on particle swarm optimization with multi fitness function is proposed. It improves accuracy and efficiency of our previous work. This proposed method consists of four parts. First, the body plane detection to identify the front and the back of the body is defined by the Eigenvector and Eigenvalue. Second, the derivative of edges of body technique is used to detect body landmarks whereby body measurements are taken. Third, the identification of the body size according to the Thai Women's sizing system performed using three main body dimensions which consist of bust, waist and hip girths. Finally, a multi-fitness function, a combination of the volumetric overlap function and the euclidean distance of body measurements ratios, is used to calculate a similarity value between 2 body data. In the final step, PSO is used to group body shapes according to each body size. Results are then compared with those obtained from k-means clustering and the identified shapes from the Female Figure Identification Technique (FFIT). The results show that our PSO clustering can dynamically group people with similar body shapes and, at the same time, avoid those that are dissimilar. This algorithm can be extend work with larger population.