|Head: Daniel Huber
Contact: Daniel Huber
Carnegie Mellon University
The 3DVIS group studies fundamental problems in three dimensional (3D) computer vision. Unlike traditional, image-based computer vision, 3D computer vision utilizes 3D sensors, such as laser scanners, flash LIDARs, structured light cameras, and stereo vision systems, to directly measure the 3D shape of the physical world.
Looking for research experience?
We are always looking for talented students who are interested in learning about 3D computer vision. Our group has the latest hardware and software technology, including several state-of-the-art laser scanners, custom built stereo vision systems, and 3D modeling software. If you are highly motivated and want to join our research group, please contact us at the email address listed above.
Our work encompasses problems ranging from low-level analysis of the limits of 3D sensors to high-level scene understanding using 3D sensor data. At the low level, we study artifacts that 3D sensors exhibit. For example, laser scanners suffer from the mixed pixel effect, which causes phantom points to appear at locations where there is no physical surface. Understanding the causes of such artifacts can help us to develop methods to compensate for them. At the intermediate level, we investigate how the low-level artifacts affect sensor performance in commonplace situations. For example, what is the effect of mixed pixels on the boundaries of scanned objects? The answer to this question is critical in the construction industry, where laser scanners are used to measure building components at a detailed level of accuracy. At the high level, we develop scene understanding methods using 3D sensors. These methods include work on object recognition, modeling, and visualization. In the area of recognition, we are interested in understanding the representations and methods necessary to reliably recognize large numbers of objects or categories of objects within 3D data. Modeling is related to recognition, but focuses more on compact, yet accurate representations. Visualization is an important aspect of 3D vision and includes human-computer and human-robot interaction.
Many fundamental problems in computer vision can benefit from the fusion of image-based methods with 3D sensing. A central part of our research involves methods for combining these two sensing modalities to achieve performance levels that are not capable with either type of sensor in isolation. Problems like obstacle detection, boundary estimation, and object tracking can all be improved with such fused information.
Our research focuses on two main application domains:
- Architecture, engineering, and construction (AEC) – 3D sensors are increasingly used to capture the geometry of buildings and infrastructure in order to create 3D models of the “as-built” or “as-is” conditions of a structure. Typically, the models are constructed manually in a time-consuming process. We are investigating ways to simplify, streamline, and (ultimately) automate this process.
- Autonomous vehicles and mobile robots – 3D sensors are heavily used in robotics applications, especially autonomous vehicles. The 3D information can greatly simplify detection of obstacles, tracking of moving objects, and estimation of vehicle position. We are investigating these and other problems within mobile robotics.
Education, Mentoring, and Outreach
Educating students at all levels is a core part of our group’s vision. We routinely mentor undergraduate researchers and visiting scholars through summer internships, independent research projects, and international exchange opportunities. We also work to educate younger students in the K-12 ages in the subject of 3D sensing and computer vision. Finally, we demonstrate our sensors and research to the community at large at open houses, technology fairs, and other events.
It is important to transform the theoretical concepts and ideas developed in academia into realizable results that can be utilized by practitioners in industry. To this end, we work with government organizations, such as the General Services Administration (GSA) and the National Institute of Standards and Technology (NIST), to translate our research results into best practices for the industry and to integrate our work into international standards. We play an active role in the ASTM International’s E57 Committee on 3D Imaging Systems, which is developing standards for laser scanners and their use in industry.