基于立体全景相机的空间几何信息提取方法研究

发布时间：2019-02-17 09:53

【摘要】：随着GIS、物联网、虚拟现实的发展,室内定位已经是“最后一米”的服务,室内环境的表达是其重中之重。全景影像(Panoramic Image)作为一种新的模型表达方式,以其独特的优势,即快速方便的数据获取在室内模型表达方面占据一席之地。全景影像还可以结合虚拟现实使人们足不出户即可漫游全世界,提供身临其境的原始场景信息。如今的全景相机大多是拍摄二维影像,二维影像虽然具有视觉上的立体感,却不具有真正的可量测性,这使得其应用范围受到很大的限制。本文从经典摄影测量理论及计算机视觉理论出发,设计了一种全新的立体全景相机设备,获得了立体全景相机周围360°×180°场景中的物点的立体影像对。基于此立体全景相机实现了全景影像的漫游及场景中空间几何信息的提取,具体研究方法和工作如下:1.设计了一套基于阵列相机的全景影像数据采集设备。共包括六对立体相机,每对立体相机由两台Gopro运动相机构成一个单元。水平方向的五对立体相机,采集水平360?的影像。朝向天空的一对立体相机,采集室内天花板影像。设备整体使用3D打印机打印的支架进行固定,最后实现了相机周围360°×180°影像的获取。2.完成了相机的标定工作。采用打印的高精度标定纸靶,利用相机标定法对相机进行标定,求得了每个相机的内方位元素和各种畸变参数。根据各个相机的畸变参数对采集到的序列影像分别进行像畸变纠正,获得了消除畸变后的序列影像。3.实现了序列影像中像点的世界坐标的获取,点位精度为±2.6cm。设备组装成功后,从摄影测量角度出发,对每一对立体相机进行相对定向与绝对定向。相对定向求得模型点坐标,绝对定向求解每一对立体相机的模型比例尺,并集成到整套系统中去。后续每次对场景拍摄一次全景。相对定向及模型点坐标获取后,乘以比例尺,就获得了物点的世界坐标,当然比例尺需要进行周期性检校。4.完成了全景影像中空间几何信息的提取。针对空间平面多边形面积、空间圆面积、坡度、角度等信息使用基于三维坐标的求解方法进行提取,利用最小二乘法进行严密平差求解。5.编程实现了立体全景相机采集的序列影像的全景漫游,并从漫游视角进行几何信息的提取。基于Harris特征提取方法对影像进行特征提取,实现了影像之间的坐标转换,采用线性融合和改善后的线性融合处理影像重叠处的色差、接缝等问题。最后以OpenGL为基础,将拼接后的影像投影到球面上,视点设置在球心,从而实现了全景影像的漫游。
[Abstract]:With the development of GIS, Internet of things and virtual reality, indoor positioning has been the last one meter service, and the expression of indoor environment is the most important. Panoramic image (Panoramic Image) as a new model representation method, with its unique advantages, that is, fast and convenient data acquisition in indoor model representation occupies a place. Panoramic images can also be combined with virtual reality to enable people to roam the world without leaving their homes and provide information of the original scene. Nowadays, most panoramic cameras shoot two-dimensional images. Although two-dimensional images have the visual sense of three-dimensional, they do not have the real measurability, which limits their application scope. Based on the classical photogrammetry theory and computer vision theory, a new stereo panoramic camera equipment is designed in this paper, and the stereo image pairs of objects in 360 掳脳 180 掳scene around the stereo panoramic camera are obtained. Based on the stereo panoramic camera, the roaming of the panoramic image and the extraction of the spatial geometric information from the scene are realized. The specific research methods and work are as follows: 1. A panoramic image data acquisition device based on array camera is designed. Each stereo camera consists of two Gopro motion cameras. Five pairs of stereoscopic cameras in horizontal direction, collecting level 360? Of the A pair of stereoscopic cameras facing the sky collect images of indoor ceilings. The device is fixed by 3D printer, and the 360 掳脳 180 掳image around the camera is acquired. 2. 2. The calibration of the camera is completed. The paper target is calibrated with high precision and the camera calibration method is used to calibrate the camera. The internal azimuth elements and various distortion parameters of each camera are obtained. According to the distortion parameters of each camera, the sequence images are corrected respectively, and the sequence images after eliminating the distortion are obtained. The world coordinates of the image points in the sequence images are obtained, and the accuracy of the points is 卤2.6 cm. After the equipment is assembled successfully, the relative orientation and absolute orientation of each pair of stereoscopic cameras are carried out from the angle of photogrammetry. The model coordinates are obtained by relative orientation, and the model scale of each pair of stereoscopic cameras is solved by absolute orientation and integrated into the whole system. Take a panoramic view of the scene at a time. Relative orientation and model point coordinate acquisition, multiplied by the scale, get the world coordinates of the object points, of course, scale needs to be periodically checked. 4. The extraction of spatial geometric information from panoramic images is completed. For the information of polygon area, circle area, slope, angle and so on, the three-dimensional coordinate method is used to extract the information, and the least square method is used to solve the strict adjustment. 5. The panoramic roaming of the sequence images collected by the stereo panoramic camera is realized, and the geometric information is extracted from the roaming angle. Based on the Harris feature extraction method, the image features are extracted, and the coordinate transformation between images is realized. The linear fusion and the improved linear fusion are used to deal with the problems of color difference and seam in the overlapped image. Finally, based on OpenGL, the spliced image is projected onto the sphere and the view point is set at the center of the sphere, thus realizing the roaming of the panoramic image.
【学位授予单位】：北京建筑大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：P235;P209

【参考文献】