动态变形表面的三维测量与跟踪方法研究

发布时间：2018-03-13 23:44

本文选题：三维测量　切入点：时空关联　出处：《复旦大学》2012年博士论文　论文类型：学位论文

【摘要】：自然场景中,有很多现象可以抽象成三维空间的动态变形运动,例如风中飘动的旗帜、水中游动鲨鱼的表面、讲话时的人脸等等。目前,尚没有成熟的手段研究这些现象中的三维运动。若能精确得到动态变形物体的三维运动数据,一方面对于研究物体的运动规律,揭示背后隐藏的机理有着重要的科学意义；另一方面存在着巨大的实际应用价值,可以大大推动相关产业的进步,如影视、动漫、游戏、工业生产等行业。为了获取动态变形物体的三维运动数据,本文提出了三种方法。第一种方法是基于联合优化的动态变形表面三维测量及跟踪方法。该方法是基于图像块的被动式测量方法,将图像划分成很多小块,同时考虑立体图像序列中的立体匹配和时间上的前、后帧之间的图像块匹配,采用时空关联的联合优化方法。优化时,假设立体图像块之间是仿射变换,而前、后帧图像块之间是平移变换。实验证明,该方法能有效地获取时空关联的动态变形表面的运动数据。然而,正是由于采用了联合优化方法,优化时,需要同时考虑较多的参数,为优化带来一定的难度。鉴于此,本文提出了第二种方法。第二种方法是从最稳定区域进行扩展的三维测量与跟踪方法。该方法提出了一种基于图像块的区域扩展技术,每次都从最稳定的区域进行扩展,通过独立的计算图像像素运动和立体匹配来进行三维的测量和跟踪。由于分开考虑立体匹配和前、后帧的匹配,优化参数减少,优化结果也更加精确,实验结果也证明了这一点。但是,该方法是一种被动式测量技术,需要被测的物体表面有较为丰富的纹理。由于存在这种局限,使用该方法测量较长的立体图像序列时,会出现累积误差问题。为了解决误差累积问题,本文提出了第三种方法,该方法采用了主动式测量技术。第三种方法是基于灰度随机三角形纹理的动态变形表面三维测量及跟踪方法。该方法需要在被测物体表面喷涂灰度随机三角形纹理。该方法中解决了如何产生随机灰度三角形纹理、如何在立体图像序列中检测灰度随机三角形、如何在立体图像中匹配三角形,以及如何在图像序列中跟踪三角形等一系列问题。实验证明该方法能精确稳定地测量动态变形表面的三维形状和三维运动,并且有效地解决了误差累计和局部光照变化问题。根据所掌握的资料,这是第一个采用灰度随机三角形纹理测量动态变形表面三维形状和运动的技术。对比实验显示,本文所提出的三种方法各有特点,但都能有效地获取动态变形物体的三维运动数据,为科学研究及工程应用提供了重要的手段。
[Abstract]:In natural scenes, there are many phenomena that can be abstracted into three dimensional dynamic deformational movements, such as flags fluttering in the wind, the surfaces of swimming sharks in the water, human faces while speaking, and so on. There is no mature method to study the three-dimensional motion of these phenomena. If the three-dimensional motion data of the dynamic deformed object can be accurately obtained, it is of great scientific significance to study the motion law of the object and reveal the hidden mechanism behind it. On the other hand, there is a huge practical application value, which can greatly promote the progress of related industries, such as film and television, animation, games, industrial production and other industries. In order to obtain the three-dimensional motion data of dynamic deformation objects, three methods are proposed in this paper. The first method is based on the joint optimization of 3D measurement and tracking method of dynamic deformed surface, which is a passive measurement method based on image block, which divides the image into many small blocks. At the same time, considering the stereo matching in the stereo image sequence and the image block matching between the pre-frame and post-frame in time, the joint optimization method of space-time correlation is adopted. In the optimization, it is assumed that the block of the stereo image is affine transformation between the blocks. The results of experiments show that the proposed method can effectively obtain the motion data of dynamically deformed surfaces associated with time and space. However, due to the joint optimization method, more parameters should be taken into account at the same time. In view of this, the second method is proposed. The second method is the 3D measurement and tracking method, which extends from the most stable region. 3D measurement and tracking are performed by independent calculation of pixel motion and stereo matching. Due to separate consideration of stereo matching and front and rear frame matching, the optimization parameters are reduced and the optimization results are more accurate. The experimental results also prove this point. However, this method is a passive measurement technique, and the object surface that needs to be measured has abundant textures. Because of this limitation, when using this method to measure long stereo image sequences, In order to solve the problem of error accumulation, this paper presents a third method, which adopts active measurement technique. The third method is dynamic deformation surface 3D measurement and tracking method based on grayscale random triangle texture. This method needs to spray grayscale random triangle texture on the measured object surface. In this method, how to produce gray random triangle texture is solved. Random grayscale triangle texture, How to detect gray random triangles in stereo image sequences, how to match triangles in stereo images, Experiments show that this method can accurately and stably measure the three-dimensional shape and motion of the dynamic deformed surface. The problem of error accumulation and local illumination variation is solved effectively. According to the available data, it is the first technique to measure the 3D shape and motion of dynamically deformed surface by using gray-scale random triangle texture. The comparative experiments show that the three methods proposed in this paper have their own characteristics, but all of them can effectively obtain the three-dimensional motion data of the dynamic deformed object, which provides an important means for scientific research and engineering application.
【学位授予单位】：复旦大学
【学位级别】：博士
【学位授予年份】：2012
【分类号】：TP391.41

【参考文献】