Improving Deep Learning Based Object Detection of Mobile Robot Vision by HSI Preprocessing Method and CycleGAN Method Under Inconsistent Illumination Conditions in Real Environment

Abstract

The family mobile robot aims to provide more intelligent services for human home life. Each sensor has its own unique features, of which the camera can be used for object detection, recognition, tracking and more. However, the effect of object detection is easily affected by the outside world such as: inconsistent illumination, object occlusion, and so on. In order to explore the effect of inconsistent illumination on object detection algorithm about robot vision in an actual home environment, the effect of illumination on object detection confidence of robot vision was studied first and the low confidence of object detection was found under the dark environment in real environment. This paper proposes a method of HSI preprocessing to improve the brightness value of HSI space to improve the effect of object detection of robot vision in dark environment. Meanwhile, specific method of brightness migration based on CycleGAN is applied to improve the detection effect of robot vision in dark environment. Compared to HSI preprocessing method, the method of brightness migration based on CycleGAN can obtain better detection effect in robot vision.