Indoor Object Distance Measurement for Robots Based on YOLO and Depth Foreground Prediction

Object detection and distance measurement are core technologies for indoor robots. RGB-D cameras enable robots to accurately calculate the distance between objects. However, the irregular surfaces of small indoor objects, including curved surfaces and protrusions, can significantly affect distance measurement accuracy. To improve the estimation of distance, an effective method for obtaining accurate object sampling region is necessary. In this paper, we propose a unified architecture to obtain accurate depth values for objects in the scene and achieve high-precision distance measurement. Firstly, the YOLO-based object detection algorithm is utilized to identify objects in the scene and obtain bounding boxes for corresponding depth images. Then, a depth foreground prediction method is proposed to predict the foreground mask assigned to the target object. Finally, a sampling method based on the depth gradient is used to select a smoother region in the foreground depth value, and accurate distance values are calculated based on this region. Experimental results demonstrate the high accuracy and practicality of the proposed method.