Repository logo
 

Improved Localization of Construction Workers in Video Frames by Integrating Detection and Tracking

Accepted version
Peer-reviewed

No Thumbnail Available

Type

Conference Object

Change log

Authors

Park, Man-Woo 

Abstract

: Tracking of construction site resources from on-site cameras using vision trackers has been proposed as a means to efficiently track a large number of objects operating in a congested environment. The systems proposed so far use first either a less accurate/faster motion detection approach or a more sophisticated object detection approach to mark the moving entities of interest in the camera view. The marked entities are then passed on to a vision tracker, who follows the entities in all subsequent frames. The major drawback of this two-step process is that the resulting tracking systems are very sensitive to changes and, in most cases, deteriorate and stop tracking effectively after only a few seconds/minutes. The reason is that it is common for construction resources to frequently a) enter/leave the field of view, b) be temporarily occluded, or c) significantly change their pose. To resolve this issue, this paper presents a hybrid method for locating construction workers that interchangeably fuses the methods of tracking and object detection. Specifically, the proposed approach continuously detects construction workers, and tracks them. Continuous detection allows a) for the detection of workers that enter the view or reappear after being occluded at any given time, b) for the effective termination of tracking when workers leave the view or become occluded, and c) for relocating the marked area and detection confidence to fit all object poses. On the other hand, continuous tracking allows for seamless tracking across time and smoothing of the temporal detection variations. The proposed approach was implemented into a C# prototype and tested on videos taken from construction sites in Atlanta, GA. The method could detect and track all workers, preserving stability under occlusions. It enhanced accuracy by 9.6% when compared to the use of a detection method only. This indicates that the proposed approach is a significant improvement over prior work and has the potential to effectively monitor the location of construction workers.

Description

Keywords

Journal Title

Conference Name

of ICSC15: The Canadian Society for Civil Engineering 5th International/11th Construction Specialty Conference

Journal ISSN

Volume Title

Publisher

Publisher DOI

Publisher URL

Relationships
Is supplemented by: