Locating Anatomical Points on Foot from 3D Point Cloud Data Jianhui Zhao 1 and Ravindra S. Goonetilleke 2 1 Computer School, Wuhan University, Wuhan, Hubei, PR China, 430072 2 Dept. IELM, Hong Kong University of Science and Technology, Clear Water Bay, Hong Kong 1 E-mail: jianhuizhao@whu.edu.cn 2 E-mail: ravindra@ust.hk Abstract Algorithms are proposed to automatically locate the foot anatomical points from scanned 3D point data based on a novel method that uses the pternion point for foot alignment, whereby variations in the different dimensions are minimized. The detected foot malleoli and arch point are used to classify the foot type. Based on the automatically detected anatomical points, 9 foot dimensions of 10 participants were determined and compared with manual measurements. 1. Introduction Foot dimensions can be used to assess the degree of fit between feet and the footwear worn [1], and can also be used to generate the 3D shapes of feet [2]. The different dimensions are based on anatomical points or landmarks, which are generally defined differently by different researchers and organizations [3-4]. Traditionally, the simpler dimensions are measured using a ruler, tape, caliper, or special devices such as the Brannock, Ritz Stick, Scholl, etc. With the development of 3D digitalization and computer techniques, automatic foot measurement is possible even though locating anatomical positions may be better performed through palpation. After the anatomical or surface points are manually determined, researchers have used various techniques to obtain the 3D coordinates of those manually determined points [5-8]. Typically, the 3D point data are then used to calculate heights, lengths, widths and angles [6,8]. Others such as Luximon et al [9] have used a limited set of landmarks to even model the shape of the foot. Point cloud, mesh or surfaces of objects can be modeled [10-11] even though the surface characteristics may not be perfect. If a set of representative points that describe the surface can be identified, the modeling errors can be minimized. However, Yahara et al [12] has shown that it is difficult to locate the anatomical points using algorithms. Some researchers [13-15] have proposed methods to detect characteristics such as edges from point cloud, mesh or surface of objects, but these are not always anatomical points on the surface of a person. Contrary to previous studies, this paper proposes a series of algorithms to locate the anatomical positions on the foot surface from 3D point cloud data without any manual intervention. The detected anatomical points are then used for aligning the foot, classifying the foot and lastly obtaining the measurements of critical dimensions that can then be used for fitting footwear. 2. Data acquisition and foot alignment The YETI foot scanner [16] was used to obtain 3D point cloud of the foot surface. The scanner manufacturer has specified the accuracy of the scanner to be ±0.5 mm. The total number of points in the point cloud depends on the length of foot, since the scan sections are set to be 1 mm apart and each section has 360 points. Dimensional data are sensitive to the reference coordinate system and hence registration is required in order to compare the data from different sources. In manual measurements, such as when using the Brannock device (www.brannock.com), the rearfoot is placed in the heel cup and the arch length pointer is slid forward so that the inside curve of the pointer matches the ball joint of the foot. Then the width measuring bar is slid firmly to touch the lateral side of the foot to uniquely locate the foot. Liu et al [6] adopted a jig consisting of two perpendicular plastic bars with three little recess holes to define the reference frame (Figure 1). Prior to digitizing, the foot is positioned such that the first metatarsal joint and centre of the heel on the medial side touches the long bar (x-axis), while the rear part of the heel touches the short bar (y-axis). In this way, the authors established an anatomical coordinate