This research aims to investigate the feasibility of a novel automated vision tracking methodology that reports the 4D location (spatial coordinates and time) of distinctly shaped, project related entities, such as construction equipment, personnel, and materials of standard sizes and shapes. Under this methodology, two or more self-calibrated, outdoor wireless video cameras are initially placed at a project site and collect video-streams. Using the construction materials and shapes visual recognition methods that the investigator has previously developed, each project related entity on the cameras' field of view is identified as an "interesting" pattern to track. The pixel area that it occupies in each camera is marked. Based on each entity's area centroid and epipolar geometry, the corresponding entities on each camera's view are matched, to determine the 4D position of each entity. Established tracking techniques are then used in each subsequent frame of the video stream to track the movement of the identified "interesting" entity while it operates within the cameras' viewing spectrum. Tracking of project related entities, such as materials, equipment and personnel, has been a significant topic of research for the last decade. Trackers, and especially automated trackers, are useful in progress monitoring and inventory control applications for construction sites, materials management, collision/accident prevention tools and security applications. Tracking records can also be used in activity sequence analysis for optimal path determination and processes redesign. Existing vision-based tracking technologies can be divided in model-based and user-driven approaches. The major advantage of model-based techniques is that they are highly efficient and precise in following the modeled targets. The limitation is their need for fairly elaborate entity models. Correspondingly, the major advantage of user-driven approaches is their wide-range applicability while their limitation is the lack of prior knowledge of the tracked entities' characteristics and the need for a user to select the "interesting" entities manually. Therefore, in a construction site where large numbers of different material types, personnel and equipment are involved, one would need to create elaborate models for each type of entity that needs to be tracked, or manually select each entity to be tracked every time that entity would enter a camera's view. These are the limitations that the proposed research addresses by automatically recognizing the related entities and directing a generic tracker to follow them accordingly.
If successfu, this research is expected to substantially increase the efficiency and quality of the information technology processes involved in civil infrastructure construction and in manufacturing. This in turn will result in cost and time savings, will improve the competitiveness of the U.S. construction industry, and will reduce life cycle costs of civil infrastructure.
