August 13, 2007
Not many students outside of drama may be able to say that they learned how to "strike a pose" at Carnegie Mellon, but last year ECE students Gary Garvin, Garrett Jenkinson, and Steven Nielsen learned how to do just that-in 3-D. The three classmates in ECE Professor David Casasent's Digital Communications and Signal Processing Systems Design course completed an assignment they titled "Strike a Pose: The 3-D Modeler," which won the David Tuma Undergraduate Laboratory Award for the most outstanding student project. Established in 1985, the award pays tribute to the late Professor David Tuma, acknowledging his dedication to undergraduate education and his commitment to practical laboratory experiences. The honor is presented at the ECE Diploma Ceremony.
"The students found an ingenious way of transforming a complicated robotic control and data processing problem into an inexpensive and efficient video processing system that produces a 3-D computer model of an object," wrote Professor Casasent his nomination letter. The Tuma award recipients are chosen from nominations submitted by faculty. "This is an excellent example of good engineering in a diverse number of areas."
In Professor Casasent's capstone design course, aided by ECE graduate student teaching assistants Wai Cheung Chu, Xinde Hu, and Rohit Patnaik, each group selected a semester-long project of its choice that had to use digital signal processing (DSP) hardware. Gary, Garrett, and Steven's goal was to construct an accurate scanner that produced 3-D models of objects at a cost of less than $100. The machinery typically used to obtain 3-D models utilizes expensive laser distance measurement technology, so many hobby engineers and independent artists cannot purchase the systems, which have applications for a variety of engineering and artistic fields, including computer-aided design (CAD), biometrics, video games, and animated movie creation.
For their final project demonstration, the teammates created 3-D models of several objects, including a shoe. They acquired a 3-D model by placing the shoe on a rotating platform and projecting a vertical laser line on it. A camera then took a picture of the shoe at an angle nearly perpendicular to that of the laser. Next, the shoe was rotated over 360Â° and images of the line were obtained for each rotation. The 3-D model of the shoe was produced from the images of the laser line, in a system that was much cheaper, faster, and simpler than standard methods.
Standard 3-D acquisition systems use data from laser rangefinders arranged around objects, requiring an intensive rotational feedback control system. In contrast, for the student's project, only the platform on which the object was placed had to be rotated to obtain the 3-D information, and they only needed one stationary camera. Additionally, the total data acquisition and processing time to rotate the object 360Â° was only 12 seconds.
"Our proof-of-concept design gave such exciting results to this effect that the countless hours of work flew by. ...In the end we were a group of diverse talents, whose relative strengths and weaknesses overlapped perfectly to make the innovative end result," Garrett recalled.
"We all enjoyed working on the project because we were taking seemingly abstract and academic theories and putting them into use in a system that anyone passing by could appreciate," said Garrett.
"This definitely provided me with the confidence that what we have learned from books and on chalkboards can be applied to make impressive real-world results," he added. "I feel much more confident testing my ideas for inventions or just sitting down to write a program which tests theories."
Casasent points out that the system has many practical features, including not having to perfectly center the object on the platform, nor rotate it about its exact geometric center. A simple motor with low speed and high torque is used to rotate the object at a constant speed. As the object is rotated over 360Â° in aspect view, a webcam snaps continuous images at fifteen frames per second, and the software then transfers the image data to a DSP board. The DSP hardware handles the image processing operations, filling in and thinning the projected image of the laser line and performing noise reduction. The hardware also performs matrix-vector operations and a coordinate transform, since there are two coordinate systems-the 3-D coordinate system of the object and the 2-D coordinate system of the image. The 3-D object is reconstructed from the 2-D images of the projected laser line profiles, using data from the angle of the laser pointer as the object rotates.
The students also developed a handy graphical user interface (GUI) through which they could change key parameters within the algorithm to optimize the performance of the system for a variety of scanning conditions. In the course of the project, the students adapted their algorithm several times using sound engineering solutions and trade-offs, including approximating some matrix-vector operations to reduce processing time and memory requirements.
Each student added their strengths to the team. "We each had a specific skill to contribute to the group, so we would generally meet to work on the project but individually focus on our part," explained Garrett, who was in charge of building much of the mathematical framework and producing results using image processing theory.
Gary was the "master programmer" who did all of the processing in real-time, making the GUI work with C code on a DSP starter kit (DSK) board. Through the DSK, he programmed the DSP to network with a standard PC.
Steven's know-how contributed to building the physical devices and making the hardware work with their software, including building the mechanical system while optimizing the laser camera system's robustness to produce results on a wide range of objects.
The group overcame numerous practical problems during the semester, including figuring out how to produce an accurate laser line projection onto an object. Their final solution used an inexpensive and common material: the laser pointed to the stem of a high quality wine glass to produce a uniform line beam.
The course assignments aimed to prepare the pupils for the issues involved in real-world engineering projects, including group dynamics, budgets, reports, and presentations. In all, Gary, Garrett, and Steven submitted a one page proposal; a written and oral proposal including a schedule, milestones, and the division of tasks between the three members; an oral update half-way through the project; a final oral presentation; a real-time laboratory demonstration on the course DSP hardware; and a written final report. All of their oral presentations were given to the entire class, allowing the professor to judge their team dynamics, oral, and writing skills.
Gary earned his B.S. last December and is working at Local Matters in Fort Lauderdale, Florida as an Associate Software Developer.
Garrett received his B.S. and M.S. last May as part of the Integrated Master's/Bachelor's (IMB) program. He is conducting global positioning system (GPS) signal analysis at Lockheed Martin in Rockville, Maryland this summer, while Steven is working at Harris Corporation in Melbourne, Florida. Steven's tasks include FPGA design for the next generation of radios for the military.
This fall, Garrett will start his Ph.D. at Johns Hopkins University where he will study imaging science for biomedical systems. Steven, who also earned his B.S. last May, will return to ECE with the IMB program, expecting to earn his M.S. this December. He is looking forward to being a teaching assistant for Digital Communications and Signal Processing Systems Design this fall.
The David Tuma Undergraduate Laboratory Award pays tribute to the late Professor David Tuma. In this photo, taken around the 1980s, David Tuma (far left) is in the laboratory with his students.
Steven Nielsen (left) and Garrett Jenkinson (center) accept the David Tuma Undergraduate Laboratory Award from Professor David Casasent. Fellow winner Gary Garvin was not present at commencement.
This picture depicts the one of the team’s results using their device—a 3-D scan of a helmet.
The group shows that the resulting digital data can be used in a CAD program for animation or analysis.
The 3-D scanner, in which the laser will point to the stem of a high quality wine glass to provide a uniform line beam.