Excavator identification method based on computer vision technology in nighttime scenes

Purpose Nighttime construction sites pose challenges for equipment detection due to insufficient lighting, with existing research often lacking applicability to nighttime environments. This paper proposes an innovative approach for excavator recognition based on computer vision technology tailored for night scenes. Design/methodology/approach The proposed approach consists of two modules: an object detection module and an image lighting enhancement module. You Only Look Once version 7 (YOLOv7) is applied as the detection algorithm in the object detection module and integrates three distinct attention mechanisms (Squeeze-and-Excitation Network, Convolutional Block Attention Module, and efficient channel attention (ECANet)) into YOLOv7 to enhance its nighttime object detection capabilities. The Night-Enhancement (NE) algorithm is utilized in the image enhancement module to boost image brightness and improve object visibility. To validate our framework, we evaluate the model using four construction videos of nighttime scenes under different lighting conditions. Findings Experimental results demonstrate that the NE algorithm effectively enhances image clarity in low-light environments, increasing the model’s average detection accuracy by 8.85%. Notably, the integration of the ECANet attention mechanism into YOLOv7 significantly enhances nighttime detection performance, achieving an average detection accuracy exceeding 97% and an average accuracy improvement of 8.63% on the test dataset. Despite the added attention mechanisms, YOLOv7 maintains a detection speed above 36 Frames Per Second (FPS) on a standard computer, enabling real-time detection. Furthermore, the object enhancement module effectively increases the visibility of targets in nighttime videos, thereby improving detection performance across most models. Research limitations/implications A key limitation of this research is that while the illumination enhancement module effectively improves visibility in low-light environments, it slightly reduces image quality in well-lit conditions. Originality/value This research presents a novel approach to nighttime excavator detection by integrating image lighting enhancement and advanced object detection algorithms.