Main Participants: Satyandra K. Gupta and Tao Peng
Sponsors: This project was sponsored by MIPS, Automated Precision Inc., and NSF
Keywords: Inspection, Reverse Engineering, and 3D Shape Measurement
Motivation
Many industrial applications require accurate and rapid measurement of the 3-D shapes of objects. Representative applications of 3-D shape measurement include reverse engineering, 3D replication, inspection and quality control. In most of these applications, users need to construct 3D point clouds that correspond to the objects surface by performing measurement on the objects surfaces. Manufacturing industry needs a fast inspection process that can measure and analyze various 3D features on the part and determine if a feature is within the tolerance specifications or not. The measurement scheme needs to be adequately accurate to eliminate measurement errors. Measurement errors can lead to erroneous inspection that results in an acceptable part being rejected and a defective part being accepted. Hence, both inspection speed and accuracy are equally important.
Coordinate measurement machines and laser based measurement techniques usually provide very accurate measurements. However, these techniques are slow because they measure various points on the part sequentially. On the other hand, camera-based techniques are usually very fast. Therefore, a possible way to perform the 3D inspection is to use digital cameras to construct a dense point cloud (e.g., points spaced less then 0.25mm apart) corresponding to the part being inspected and then analyze the point cloud to determine if it meets the tolerance specifications. But accuracy associated with the conventional camera based inspection techniques has not been very high in the area of measurement of geometrically complex 3D shapes.
Shape measurement based on digital fringe projection (SMDFP) is a technique for non-contact shape measurement. Due to its fast speed, flexibility, low cost and potentially high accuracy, SMDFP has shown great promise in 3-D shape measurement, especially for applications that require acquisition of dense point clouds. A typical SMDFP system contains one projection unit and one or more cameras. During the shape measurement process, a set of fringe patterns, whose structures are accurately controlled by computer, are projected onto the surface of the object being measured. Meanwhile, the images of the object shone by the light patterns are captured by the digital camera(s). By using image processing techniques and some variation of a triangulation method, a dense 3-D point cloud representing the surface of the object can be constructed.
We are interested in developing a comprehensive mathematical model for SMDFP and the associated shape measurement algorithms.
Main Results and Their Anticipated Impact
SMDFP system being used in our research utilizes a digital micro-mirror device (DMD) to generate a projection pattern and digital camera to take the images. This system generates an appropriate projection pattern and uses a DMD-based projection unit to project the pattern on the object being measured. The digital camera takes images of the object. Due to three-dimensional nature of the object surface, the projected pattern distorts. The images captured by camera records the distortion in the projection pattern. Images captured by the camera are analyzed by the system to estimate the 3D points on the object surface that cause the distortion in the projection pattern seen in the image. The system finally returns a 3D point cloud that represents the object surface. DMD-based projection unit provides excellent resolution and brightness, high contrast and color fidelity, and fast response times.
One of the key innovations behind our system is use of multiple projection patterns. Different projection patterns lead to different accuracy. A projection pattern that produces accurate result for one shape feature may not be ideal for some other feature. Hence, different projection patterns are needed to capture different features on the object accurately. The use of multiple projection patterns allows the new system to measure all the features on the object accurately. The system also selects the projection patterns carefully to minimize the number of patterns being used to keep the measurement process fast. Another novel feature of the system is use of a high fidelity mathematical model for every element of the system. This helps in improving the overall measurement accuracy.
Our main results include:
- Developed detailed mathematical models and functional structure of the shape measurement system.
- Developed algorithms to estimate various system parameters.
- Developed algorithm to generate dense point clouds by analyzing images.
- Developed procedures for determining the appropriate projection patterns.
- Developed prototype shape measurement software.
This software generates point clouds and provides visualization capabilities to examine the generated point clouds. This system seems to work very well for a wide variety of shapes. A noteworthy feature of the system is that it works extremely well with parts with holes and discontinuities. These kind of parts posed tremendous difficulties for vision-based technologies in past. Even for complex parts, the system only needs to take eight images to produce very good results. Hence, it is a very fast system. API has also done evaluation of accuracy achieved by the system. On the test parts supplied by Ford, the system produced average error of less than 75 microns. We believe that with some fine-tuning we will be able to reduce this error to below 50 microns level.
API plans to release a commercial called 3D Rapid Scan based on our research results. In summary, we have developed one of a kind shape measurement system that generates dense point clouds with unprecedented speed and accuracy for a wide variety of complex parts. This system forms a basis for performing cheap, fast, and accurate 3D inspection and has the potential for opening new markets for API.
Related Publications
The following papers provide more details on the above-described results.
- T. Peng, S.K. Gupta, and K. Lau. Algorithms for constructing 3-D point clouds using multiple digital fringe projection patterns. CAD Conference, Bangkok, Thailand, June 2005.
- T. Peng and S.K. Gupta. Model and algorithms for point cloud construction using digital projection patterns. ASME Journal of Computing and Information Science in Engineering, 7(4): 372-381, 2007.
- T. Peng and S.K. Gupta. Algorithms for generating adaptive projection patterns for 3-D shape measurement. ASME Journal of Computing and Information Science in Engineering, 8(3), 2008.
Some of these papers are available at the publications section of the website.
Contact
For additional information and to obtain copies of the above papers please contact:
Dr. Satyandra K. Gupta
Viterbi School of Engineering
University of Southern California
Los Angeles, California 90089-1453
Phone: 213-740-0491
Email: guptask [AT] usc [DOT] edu