Object Location Research For Binocular Stereo Vision Based On The SIFT Algorithm

With the rapid development of the science and technology, the Robots are widely used in the areas of people’s production and living. Many researchers are full of enthusiasm in robotics. As the key technology in the robot, the robot vision is the key and difficult point for people to study. Now, typically the robot vision is the binocular stereo vision technology. The technology get the object’s information from the outside world by simulating human’s eyes, which had been used in military and civil fields. The binocular stereo vision technology uses get objects from different angles by the dual cameras, and establish its mathematical model, Then get3D information of object according to the calculated. Usually a complete vision system has several steps, such as the Images’acquisition, the camera’s calibration, extracting and matching feature points, Calculating coordinates and recovering3d information. The camera calibration and image feature point matching is difficult point. As many methods of camera calibration, the different methods affect the precision of calibration, the algorithms of image matching are many, the different matching algorithm affects the speed and precision, The paper mainly studies the SIFT feature extraction algorithm. The choose of the camera calibration method and image processing algorithm has impacted on the real-time and accuracy of the system. This paper has made an in-depth research.The principle of binocular stereo vision and the mathematical model are introduced in this paper, and the internal and external parameters of camera need calibrated. Although the traditional method of the camera’s calibration is more precise, but need more time-consuming. If the environmental changes, it is needed to adjust the camera focal length all the time. In the case of inaccurate on the self-calibration, the article chooses the methods between traditional and self-calibration method——Zhang Zhengyou calibration method. The method is based on multiple image plane template for calibration, through the Angle of point check measure the feature point on the template image coordinates. Then through the experiment and calculation, the camera parameters can be obtained.The images are got from the binocular camera are needed to match. This paper analysis the characteristics of the SIFT algorithm is introduced and the extraction process. This is a kind of algorithm based on scale space, the image zooming rotation and affine transformation keep invariance. In the matching feature point detection, also has good noise resistance. But the algorithm also has some disadvantages, such as excessive detected feature points, the extraction of feature points need large amount of calculation descriptor and the larger dimensions. Therefore this paper discusses some improved algorithms, such as PCA-SIFT algorithm, SURF algorithm and Harris-SIFT algorithm. And these algorithms are compared with those of the image matching SIFT algorithm, Indeed the time of extracting feature points consumed and the accuracy of the feature points has improved a lot.Finally, the paper has carried on the experiment of binocular stereovision. First of all camera is calibrated, Then the images are matched simulated on the matlab software on the basis of the SIFT algorithm. the3D information is calculated according to the object image coordinates. After using the improved Harris-SIFT algorithm, the real-time and accuracy is greatly improved.