A Case Study of Construction Equipment Recognition from Time-Lapse Site Videos under Low Ambient Illuminations

摘要

High definition construction cameras have been increasingly placed on construction sites to record jobsite activities into time-lapse videos. These videos have been used in several research studies to promote construction automation related to productivity analysis, site safety monitoring, etc. However, most videos tested in these studies were collected at daytime. It is not clear whether the time-lapse site videos collected at night are still useful, considering construction work might be performed both day and night for various reasons. The main objective of this paper is to investigate the effectiveness of recording jobsite activities with the time-lapse videos collected by construction cameras at night or other low ambient illuminant conditions through a case study. The construction site of the Romaine Complex project in Quebec is selected as the test bed. The cameras have been placed on the site to record the jobsite activities from day to night. All the collected site videos are classified based on their ambient illuminant conditions. Then, the videos under low ambient illuminations were tested with the object recognition techniques that have been commonly used in existing construction research studies. The recognition results indicated that the videos collected at night or other low ambient illuminant conditions are still useful and could be considered as an important source for site data sensing and analysis.