| The blind are the disadvantage community in our society, how to improve thequality of life for blind people, so that they can be more independent and autonomousin daily life and work, has become an urgent social problem to solve. From thepractical needs of indoor blind assistance, this thesis, which was based on computervision technology, did a profound research about the indoor blind assistance system.The research and experimental tests were focused on indoor objects orientation andtargets identification.Binocular stereo vision is one of the important research directions in computervision technology, which is based on the principle of parallax that similar to thehuman eye vision. Binocular stereo vision technology generally taking two images ofthe target objects by two digital cameras from different angles at the same time (Thismethod can be used for the measurement of moving targets), or by one digital camerafrom different angles and at different times (This method only applies to stationarytargets), and with the stereo disparity and triangulation, the3-D geometricinformation of the target can be recovered. It is easy to obtain the distance betweenthe camera and the target, realizing positioning by using the binocular stereo visiontechnology. Based on binocular stereo vision technology, this paper was aimed at thestudy of target location and recognition, and obtained some preliminary results.Firstly, the binocular camera clibration and target location were implemented. Inorder to recover the three-dimensional information of objects in space from thetwo-dimensional image that photographed by the camera in binocular stereo visionapplications, it is have to calibrate the camera, and after calibration the mappingrelationship between the object space and computer image space could be established.In this paper, the transform relationship in the processing of camera calibrationbetween the world coordinate, camera coordinate, image plane coordinate and pixelplane coordinate have been analyzed. The camera was modeled by the usual pinholemodel, and both the radial and tangential distortion that generated by the optical lenswere considered, and established the mathematic correction model. By using the planepattern method of Zhengyou Zhan and the OpenCV library, the binocular cameracalibration was realized. On the basis of calibration results and based on the epipolargeometry principle, the image distortion and corrected the stereo images wererectified. And the image stereo matching was completed by taking advantage of SAD (Sum of Absolute Difference) method and epipolar constraint.Secondly, the target recognition that based on QR Code label was implemented.Beyond the positioning function in the blind aid system, the function of targetidentification is also needed. But the traditional target identification algorithms isdifficult to identify all target objects exactly, and the results may not meet the needs ofthe blind, as the indoor environment is complex, there are a wide range of objectswith different shapes and be placed clutter. So a recognition method based on QRCode label was put forward in this paper. The designed QR Code label was with a redcircular pattern as the outside mark, it is easy to separate the red sign from the blackand white QR Code according to the color characteristic, and the color characteristicsnearly does not rely on the size, the direction, the viewing angle of the image, also ithave invariant feature for translation and rotation transform, it with strong robustness.The label designed with circular boundary, so it is able to locate the center positionand the radius of the QR Code accurately by detecting the outside mark. And thecircular mark is not sensitive to noise and light, so that the stability of the outsidemark detection will be better, that can improve the efficiency of label identification. |
PDF Full Text Request