Comparative Study of Human Skin Detection Using Object Detection Based on Transfer Learning

With the increasing aging of the population, the design of automatic bath robot has the forward-looking significance. The robot needs to detect the skin position, so as to perform the bathing task. The perception of skin is the key technology to achieve the bathing task. In this paper, object detection is used to identify the skin, which provides reference information for the pose of the robot. According to the classification of the object detection algorithms, this paper selects four typical object detection algorithms, namely, Faster R-CNN, YOLOv3, YOLOv4 and CenterNet. Due to the limitation of the self-built data set, this paper adopts the transfer learning to promote the completion of new tasks, which takes the pre-trained model as the starting point. The experimental results show that the detection results of YOLOv4 is the best, with mAP of 78%. This paper proves the feasibility and effectiveness of object detection completing the human skin detection in the bathing task.