| description abstract | Control systems have proven to be beneficial in improving the productivity of earthmoving operations. A main component of these systems is the monitoring module. Computer vision algorithms are among the new methods that have been tested for real-time monitoring of earthwork activities. These methods, however, were able to detect only intraclass equipment and failed to identify individual machines, which is a key disadvantage compared to radio-based devices, namely global positioning systems (GPS). To address this issue, a pipeline framework, consisting of several computer vision algorithms, has been developed to identify individual machines. In this framework, an object detection method is used to locate construction equipment. If a detection view of a target is obtained, the camera zooms on the candidate to identify visual markers attached on the machine. The architecture of this system is optimized by employing time-consuming processes only for the most probable candidates. This system was evaluated using several real-time videos, and demonstrated promising performance in identifying excavators and dump trucks, with 89 and 84% identification rates and 64.6 and 77.1% recall rates, respectively. In addition, applying the marker-based verification step proved to be effective in rejecting false alarms as the precision was 100% in both test cases. | |
