Design of library intelligent lighting system based on deep learning

Area of Science:

• Computer Vision
• Artificial Intelligence
• Smart Systems Engineering

Background:

• Current university library lighting systems suffer from low intelligence and high energy consumption.
• There is a need for automated and adaptive lighting solutions to improve user experience and reduce energy waste.

Purpose of the Study:

• To design and implement an intelligent lighting solution for university libraries based on sub-area autonomy control.
• To enhance human target detection and spatial localization accuracy for adaptive lighting.

Main Methods:

• An improved YOLOv5 algorithm incorporating a coordinate attention mechanism and depth-separable convolution for human target detection.
• A multilayer perceptron (MLP) model for mapping 2D human coordinates to 3D spatial localization.
• Optimization of the bounding box regression loss function in YOLOv5.

Main Results:

• Achieved a mean average precision (mAP) of 91.4% for human body detection.
• Demonstrated fast and accurate localization of multiple people within the library space.
• Experimental results showed a 6.9% point increase in mean average accuracy compared to baseline methods.

Conclusions:

• The proposed intelligent lighting system effectively performs human body detection and localization.
• The system offers a practical and referenceable solution for developing intelligent lighting in academic libraries.
• This approach significantly improves the intelligence and energy efficiency of library lighting systems.