With the availability of inexpensive Kinect depth cameras, questions arise about accuracy of depth measurements from these cameras for scientific purposes. We quantified the uncertainty of depth values within the dynamic range of Kinect camera.
The calculation of geodesic measurements has a wide range of applications, including applications in the medical field, material science field and machine vision field. We aimed at quantifying the accuracy of geodesic measurements (distance and area) on a variety of objects including rigid, deformable, articulated and non-articulated objects.
A Lego model of a calibration object was constructed with two parallel flat surfaces that were separated by 64 mm (see Figure 1). The calibration process consists of moving the Lego model away from the Kinect along the depth axis, correcting for its tilt, detecting the frontal and posterior parallel planes, and plotting the difference between detected planes against the measured depth value. Figure 2 shows the measurements over a range of 0.5m to 3.5m.
In addition, we focused on developing software that would assist in measuring geodesic distance and area of human body parts using Kinect camera. The application scenarios include tracking geodesic measurements on a moving deformable object, particularly a human, and measuring geodesic distance and area on a three dimensional model of a rotating rigid object. These application scenarios need components for spatial detection, object tracking, coordinate transformation, and three dimensional reconstruction.