Abstract

Smart lighting systems can reduce energy consumption by controlling lights according to the distribution of personnel. Most of the current smart lighting systems using vision-based indoor personnel positioning method suffer from difficulties in calibration and deployment. A smart lighting control system based on fusion of monocular depth estimation and multi-object detection is proposed in this paper. The system automatically recognizes the relative positions of people and lights from the video and controls the lights, thereby reducing the complexity of calibration and deployment. In order detect the relative positions, a multi-object relative position detection algorithm (Deep-YOLO) is proposed. Deep-YOLO fuses and optimizes Scaled-YOLOv4 and AdaBins, and it detects the relative positions of people and lights by performing simultaneous multi-object detection and depth estimation on videos. A dedicated image dataset (PL) and deep transfer learning are utilized to improve the object detection accuracy of Deep-YOLO. In order to control lights according to the relative positions, a K-Means-based lighting control algorithm is proposed, and a smart lighting control system verification platform is developed. The experimental results show that the mAP of Deep-YOLO is 98.16% and the control average error of the system is less than 1 m.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call