4-Dimensional Measurement System - 3D Measurement of Anatomocal Cross-sections of Foot during Walking -
Most of the existing systems for measuring 3D shapes use time-multiplexed images with controlled light sources (e.g. laser scanning, switching pattern projection). It is difficult for those systems to measure the 3D deformation or motion. The purpose of this research is to measure the changes of the foot shape during walking. Our desired accuracy for practical purpose is less than 1mm error. The process does not have to be real-time.
In the field of Computer Vision, 3D reconstruction is a common challenge. Many researches about stereo-matching system have been conducted. Prof. Steve Seitz et al. compare the recent multi-view stereo methods in their project. Even though their result seem to be fine, their accuracy is not enough for our purpose. In consideration of the target's scale, their best method's error is equal to approx. 2 to 3mm in measurement of foot shape.
In this research, we measure the anatomical cross-sections of foot during walking. We are not aim to propose general stereo-matching methods. The target cross-sections are painted in individual colors. Using the synchronized video cameras, our specialized multi-view stereo method acquires 3D shapes in frame-by-frame. For evaluating the accuracy, we measured the shape of a plaster foot in motion imitating human walking. In comparison with the measurement by an existing (static) 3D scanner, the error was always approx. 0.2 to 0.3mm.
The 4 lines in the figure are cross-sections defined by anatomical landmarks. In Japanese Industrial Standard (JIS), shoe size is defined by 3 parameters; "length of foot", "perimeter of Ball", and "width of Ball".
- Camera calibration
- Accurate detection of the calibration points for Tsai's method.
- Improvement of calibration using bundle adjustement
- Selection of camera pairs for stereo matching
- Selection based on the direction of cameras
- Selection based on the angle between epipolar plane and cross-sections
We constructed 1m height and 10m length catwalk, which is surrounded by 12 shutter-synchronized IEEE1394 cameras. Each camera is connected to a Windows PC. The capturing program is remotely controlled via TCP/IP. This system captures multi-view stereo movies at 14FPS (70msec interval). The captured images are XGA (1027x768) size of 16bpp (4 Bytes/2 pixels) YUV pixels.
- "3D Measurement of Feature Cross-sections of Foot while Walking", Makoto Kimura, Masaaki Mochimaru, Takeo Kanade, Machine Vision and Applications (to be published)