Distance estimation is essential in developing humanoid soccer robots. Accurate distance measurement can minimize an error while the robot is maneuvering, chasing a ball, or passing the ball to the proponent robots. Currently, stereo vision and feature matching is the conventional method to estimate the distance. Distance is estimated based on the disparity value between detected features on the stereo image. However, the matching process needs high cost computationally. Furthermore, the estimated distance based on feature matching is less accurate. Therefore, in this work, the distance estimation based on the object coordinates detected using the YOLOv3 has been proposed. Additionally, a linear regression algorithm added to improve the measurement accuracy. Several experiments have been done to verify this proposed method in realtime applications. As a result, our proposed method successfully improves the distance measurement accuracy from 86.58% to 98.01%.