Research Visual Localization Method For Vehicle-based Image Sequence Of Along The Optical Axis

In recent years, because of rich environment information collection, low price and easy touse, visual sensor was subjected to universal concern, based on visual sensor orientationmethod has become the focus of research, and monocular vision more because of its simplestructure location and easy to implement and a wide range of applications, but also avoid thebinocular or more eye in the vision of small, three-dimensional view the disadvantages ofmatching difficult[34]. This paper mainly based on car along the optical axis sequence imageson the visual position method by using sequence images of motion platforms will realizepositioning posture, from sequence images to the ground in target detection, and carry on thetrack and3d position. This paper research content mainly includes the following aspects:1. The general digital camera calibration. Digital camera parameters are generallyprovided by the manufacturer, because in the visual position is not fixed in the focal length, andthe general digital camera lens distortion is compared commonly big, so in practical applicationmust coat internal position before for calibration. In this paper, according to the principle of thevanishing point of real-time camera calibration, to avoid the fixed focal length restrictions, andwill RANSAC (random sampling uniformity) use the vanishing point camera calibration,improve the calibration robustness.2. Along the optical axis sequence images research problems. Aerial photography in themeasurement of rotation Angle of general is small, so the resection of initial value can assumeinitial value is0, however along the optical axis rotation Angle of the image is very big, theconventional space rendezvous methods have been to fit in, so this paper using quaternionsmethod for spatial resection.3. Sequence images of the same point extraction and image matching. The image ofseveral existing feature point extraction algorithm compared, and the method of selecting SIFTas this research method, of the detail, finally achieved the point and the extraction of imagematching.4. Kalman filter principle and the application in visual position. As in the anterior chamberfair-resection continuous process, and on the basis of error propagation theory, the error will gradually accumulative total, so this paper using kalman filtering, for each station successivefiltering, improve the precision of the positioning.