A Salient Region Detection Method Based on Frequency-Tuning and One-Cut Algorithm

Domestic robots have shown up in more and more families in recent years. When they follow the order from the host and serve a family, it is involved in many techniques such as object recognition, object grasping and so on. So target segmentation is a particularly important technology, which is also a very challenging problem in the field of image analysis and computer vision. Based on the previous work, we can see that the Frequency- Tuned salient region detection (FT) is superior in performance and computational efficiency and can meet the realtime requirements of the robot system. However, the output result is a saliency map in pixels at full resolution, which contains no “object“ concept. In order to extract the complete salient region, we fuse FT algorithm and One-Cut segmentation method to achieve automatic segmentation of salient goals. In addition, the accuracy and robustness of saliency region segmentation are improved by using connected component attributes and depth gradient features.

I. Introduction

Due to population aging [1] in recent years, the request for domestic robots has been raising and they will play a significant role in our daily life in the future. Domestic robots usually have multiple sensors installed for multimodality sensing. Vision is the main source of human perception of the world. Human visual system has been highly developed and perfected and they obtain 80% of external information through visual access. In order to perceive the surroundings as human beings do, domestic robots often carry vision sensors like cameras for rich scene information, but at the same time there are information redundancy issues.