pcl-Overview-three-dimensional image processing

Point cloud model and three-dimensional information

Three-dimensional image is a special form of information expression, which is characterized by the data of three dimensions in the space of expression. Compared with the two-dimensional image, the three-dimensional image can realize the natural object-background decoupling by using the information of the third dimension. In addition, for the visual measurement, the two-dimensional information of the object tends to change with the projective mode, but its three-dimensional features have better uniformity to different measurement methods. Unlike photographs, three-dimensional images of a class of information collectively, the information needs to have a specific form of expression. Its manifestations include: depth mapping (the distance between objects and cameras in grayscale), geometric models (built by CAD software), point cloud models (all reverse engineering devices sample objects into point clouds). Visible, point cloud data is the most common and most basic three-dimensional model. Point cloud model is often measured directly, each point corresponding to a measurement point, without any other means of processing, it contains the largest amount of information. However, the information hidden in the point cloud needs to be extracted by other means, and the process of extracting information from the point cloud is a three-dimensional image processing.

Three levels of point cloud processing

Like image processing, point cloud processing also has different levels of processing. Or, according to the requirements of the task, different processing methods need to be combined, and these processes have a succession of points. Marr the image processing into three levels, including image enhancement, filtering, edge detection and other basic operations. The middle level includes the connected domain tag (label), image segmentation and other operations. High level includes object recognition, scene analysis and other operations. The tasks in the project often need to use multiple levels of image processing methods, in the traditional image processing method (the traditional is not including CNN neural network and large data set), the process of image processing need to use different levels of image processing to complete the task.

PCL official website on the Point cloud processing method gives a relatively clear level of division,.

The common here refers to the type of point cloud data, including many types of point clouds such as XYZ,XYZC,XYZN,XYZG, and ultimately, the most important information is contained in the POINT<PCL::p oint::xyz>. As can be seen, low-level point cloud processing mainly includes filtering (filters), key points (keypoints), Segmentation (segmention). The filtering, edge detection and segmentation are respectively corresponding to the image processing. Obviously, in the image processing is still the middle Level division operation, because the point cloud characteristic is simplified to the low level, in essence with the filtering and the key point extraction difficulty is quite. The mid-level processing of point clouds is a feature description (feature). High-level processing includes registration (registration), identification (recognition). It can be seen that point cloud is more advantageous than image processing in the difficulty of segmentation. Accurate segmentation also provides a good basis for recognition.

pcl-Overview-three-dimensional image processing