Deep learning-based human head detection and extraction for robotic portrait drawing

摘要

This paper presents a head detection and extraction method that can be used in robotic portrait drawing. First, using the state-of-the-art, real-time object detection system-YOLO(You Only Look Once), we train the model to automatically detect human heads directly from the image. Then we utilize the holistically-nested edge detection (HED) algorithm to extract head edges by performing image-to-image prediction. Finally, the content image of the head can be synthesized into a head edge map with a style synthesis algorithm. The synthesized image can be sent to the robot for drawing. Our method was verified and evaluated on the drawing robot we developed.