Real-time laser projection and deep learning for dynamic object detection and sorting assistance in industrial applications

In this paper, we propose an end-to-end perception-to-projection laser-assisted collaborative sorting (LACS) and a dynamic object detection system that continuously converts real-time detections of moving objects into target-specific physical laser cues for human operators. The proposed method integrates object detection, laser galvanometer-based visual projection, and spatial calibration into a unified perception-to-projection architecture to enable efficient human–machine collaboration in dynamic industrial environments. Besides, the YOLOv11 model is employed to accurately detect small and occluded objects, while a homography-based mapping method enables precise alignment between the imaging system and the projection module. To support real-time interaction, a lightweight laser projection module with a custom galvanometer control circuit is developed, achieving a refresh rate of 118 Hz and rendering up to 936 points per second. Experimental results show that the projection error remains below 2%, corresponding to less than 2 cm within a 1 m working range. The system is validated in two real-world scenarios: (1) a conveyor-based fruit and vegetable sorting task, where projected cues guide human operators with an average detection accuracy of 97.18% at a conveyor speed of 0.6 m/s; and (2) a precision lens assembly task, where projected markers assist manual positioning and verification. The results demonstrate that the proposed system effectively enhances human task execution by improving perceptual clarity, reducing operational ambiguity, and enabling efficient human–machine collaboration in real-time industrial environments. Therefore, our proposed system improves efficiency, reduces errors, and ensures robust, fast, and simple operation across industrial applications.