| |
Objective:
Provide fundamental measurement science, tools, and testbeds to facilitate industry development and deployment of construction object recognition and tracking systems.
Problem:
Maintaining situational awareness on a cluttered and dynamic construction site is a challenging problem. Improved knowledge of the location and movement of tradeworkers, construction equipment, and manufactured construction components is an important enabler for site safety, security, and productivity. Automatic determination of this information from a future intelligent site (e.g., employing camera networks, RFID/UWB/GPS-based asset tracking systems, laser-based 3D imaging systems) would not only enable both lean construction and enhanced security, but would also lay the foundation for levels of construction automation envisioned in the FIATECH Intelligent and Automated Construction Job Site (IAJCS). The FIATECH IACJS road-mapping initiative identified approximately twenty-five critical research areas as ranked by leading industry representatives from Dow, Black & Veatch, Fluor, Intel, GSA, CH2MHill, and others. Construction object recognition and tracking is a component of at least ten of these research areas.
Although there have been research efforts in construction-related computer vision and increased emphasis on the use of auto-identification systems on construction sites (e.g. barcode, RFID), the problem is not solved and remains an active research area. This is a particularly difficult problem given the wide variation in environmental conditions, the rate at which construction sites change over time, the amount of clutter in which a system would have to operate, and the wide variation in appearance of construction site objects. Construction object recognition as a NIST research area is well-founded given our expertise in 3D imaging systems, tracking technologies, and construction object information models. This work also provides foundational capabilities for related projects in Construction Control Using 3D Imaging and Building Information Models and The Intelligent and Automated Job Site Research Testbed.
Approach:
The approach for this project is to use feedback from sensors that could provide construction object tracking (e.g., GPS, UWB, RFID) to seed model-based recognition using camera networks and active emission 3D imaging systems (e.g. laser scanners, range cameras). This work builds upon prior efforts in laser-based tracking research (ComTrack) as well as on-going efforts in information modeling and developing algorithms for registering and analyzing 3D image data.
A related result of this work will be the development of a standard testing paradigm in which future construction object recognition algorithms and technologies can be compared.
In FY08, BFRL researchers will develop and employ probabilistic sensor models of high frame-rate range cameras for use in model-based construction object recognition and tracking systems. These sensor models will first be used to develop synthetic data sets of target objects with levels of sensor noise that vary from a perfect sensor to measured sensor performance. These data sets will be used to develop and test algorithms for model-based recognition systems. Also in FY08, initial work will be conducted to combine standard and range camera images to enhance object segmentation and identification. Finally, initial research will be conducted on the use of calibrated camera networks (e.g., a series of registered web-cams which ring a construction site) for object tracking. Two target demonstrations will take place in FY08 in conjunction with the IACJS Research Testbed. The first will be the autonomous docking of steel beams by the NIST RoboCrane using on-board range cameras. The second will be the automatic tracking of personnel using a calibrated camera network.
In FY09, the construction object recognition algorithms will be optimized and extended to include objects within a large data base of steel beams derived from a Building Information Model (BIM) and initial work will take place on recognizing, locating, and tracking construction equipment as well as people with the camera networks.
In FY10, the project will conclude with work to combine multiple sensor modalities such as camera networks, installed 3D imaging systems, mobile 3D imaging systems, and RFID/UWB/GPS systems.
|
