Sub-pixel Distance Measurement Algorithm Based on Improved SURF
Abstract
A sub-pixel-level measurement algorithm based on improved SURF is proposed. Firstly, the sobel operator is used to extract the edge of the image quickly in order to highlight the potential feature point, and reduce the detected range of the SURF algorithm. Then use the SURF algorithm to obtain the sub-pixel level coordinates of the matching feature points in the two images. After the match, the affine transformation remove the wrong match point. Finally, the moving distance is calculated by using the sub-pixel-level coordinates of the matching points. The experimental results show that the error of the moving distance is less than 3 %, according to the feature points of the two images. And if the original image resolution is 2560 * 1440, the required running time is only 0.3 to 0.4 seconds.
Keywords
Download Options
Introduction
At present, there are many image-based measurement distance algorithms have been proposed. These image-based ranging methods can be roughly divided into two categories. In the first categories, the image obtained by a single camera is processed to measure distance. For example, Suraphol proposed to measure the length of an object [3] by using the acceleration signals and object differences. Han et al. used the highest feature point of the marker crown and the triangular similarity theory to measure the length [4]. Zhao et al. proposed to obtain the centroid of the odd and even field by using the interval scan to measure speed [5]. The second category is through two cameras to obtain images, using the difference between the two images to calculate the distance. For example, Yasir et al. used visual differences and camera focal lengths from two parallel cameras to measure distance [6]. Hou A-Lin built stereoscopic vision with two cameras and then calibrates with the projection matrix to measure the safe distance traveled by the vehicle [7]. Hai-Sung Baek proposed to use two cameras to build stereoscopic vision. Through the difference between the two images calculated in a larger field of view, a more accurate measurement of object distance is got [8].
The first categories are not only the image processing, in most cases also need to utilize laser, ultrasonic and other auxiliary equipment to obtain relevant information. The second categories require multiple devices to coordinately use, with more complex hardware devices. Thus, a new distance measurement based on single image is proposed in this paper, which is a more economical and convenient measurement manner.
The SURF algorithm is generally used for matching, comparing and tracking of target objects in various scenarios. Its advantage is that it can quickly locate the coordinates of the feature points, and the coordinates can be subpixel [11-14]. In this paper, the Sobel edge extraction is utilized to reduce the scope of feature extractions, and affine transformation is utilized to avoid false matching result of the SURF algorithm. Afterwards, the SURF algorithm will be used to extract the feature points before and after the target object is moved. Then the result of feature extractions is calibrated by affine transformation. Combined with the angle information, calculate the average of the components from multiple groups measured length values in the direction of motion, and utilize this average value to estimate the distance. The innovation of this paper is to apply the SURF matching algorithm to distance measurement. It not only can quickly and stably obtain the result of the measurement, but also the coordinates of the measured feature point which can reach the sub-pixel precision.
Conclusion
This paper presents a subpixel distance measurement algorithm. It has rotational invariance, scale invariance, and satisfies the velocity measurement in strong light and low light. The experimental results show that according to the feature points of the image, the error between the measured distance and the actual moving distance is less than 3 % and the time is only 0.3 to 0.4 seconds. The measurement results are retained to 0.0001. The algorithm is applied to the measurement of distance, and can also be used to measure the speed. The algorithm can be used for the speed measurement of the moving objects in a quiescent state. It can also be made on the vehicle equipment, to measure its speed.