基于全景环带立体成像系统的深度信息估计

Depth Estimation with a Panoramic Stereo Imaging System

提出了一种基于全景环形透镜(PAL)的全景环带立体成像系统的立体信息提取方法。该成像系统由两组共轴PAL单元构成,其成像圆为两内外相接的圆环,能够实时提取360°水平视场角物体的立体信息。基于该系统的成像原理,针对非单视点成像系统特点建立相机模型对系统进行标定,并验证了标定结果。将尺度不变特征变换(SIFT)算法应用到对应点匹配上,充分利用本系统的共线约束来提高匹配的速度和准确度。应用三角测量原理从捕获的图像中提取有效深度信息。进行立体信息提取实验,给出了结果和误差分析。该系统能有效获取场景深度信息,3m距离范围误差率在±5.2%内,证明了该系统的可行性。

An information extraction method based on a panoramic stereo imaging system with double coaxial panoramic annular lens （PAL） units is presented. Its imaging circles are concentric so that depth information of 360° horizontal field-of-view （FOV） object can be extracted in real time. Based on imaging principles, the system is calibrated with a model that can accommodate the characteristics of a non-single viewpoint system, and the calibration results are verified. The scale-invariant feature transform （SIFT） algorithm is applied to the corresponding point matching, and it makes full use of collinear constraint of the system to improve the matching speed and accuracy. Effective depth information is extracted from the captured image with applying triangulation principle. Depth information extraction experiments are performed, and the results and error analysis are given. This system can restore depth information effectively, and its error rate of 3-m distance range is within ±5.2%, which proves the feasibility of this system.