Idle Identification of Construction Machinery through a Deep Learning-Based Algorithm Embedded in Surveillance Camera Systems

This study proposes a lightweight deep learning-based algorithm for idle identification of construction machinery, which can be embedded in surveillance camera systems. The construction industry faces severe challenges. Efficient utilization of construction machinery is crucial. Monitoring the utilization rate of construction machinery can identify inefficiencies and idle times, allowing for optimization of equipment use. The proposed algorithm consists of an object detection model, tracking algorithm, and idle state identification method. It is designed to run on a CPU and on the edge. The embedding in the existing surveillance infrastructure has several advantages, such as leveraging existing hardware to reduce costs and minimizing bandwidth usage and latency by enabling edge deployment. Performance findings indicate that the algorithm can effectively monitor idle states of construction machinery, achieving high accuracy and a high harmonic mean of precision and recall overall.