1. Monocular vision system
Monocular vision systems use only one vision sensor. A major drawback of monocular vision systems is the loss of depth information during the imaging process, as they project an N-dimensional image from the three-dimensional objective world. Nevertheless, due to their simple structure, mature algorithms, and relatively low computational cost, monocular vision systems are widely used in autonomous mobile robots, such as for target tracking and indoor positioning and navigation based on monocular features. Furthermore, monocular vision forms the basis for other types of vision systems, such as binocular stereo vision and multi-view vision, which are achieved by adding other methods and measures to monocular vision systems.
2. Binocular stereoscopic vision system
Binocular vision systems consist of two cameras that use triangulation to obtain depth information of a scene. They can also reconstruct the 3D shape and position of surrounding objects, similar to the stereoscopic function of the human eye, and the principle is simple. Binocular vision systems require explicit knowledge of the spatial relationship between the two cameras. Furthermore, obtaining the 3D information of the scene environment requires two images of the same scene captured simultaneously by the two cameras from different angles, and complex matching is necessary for accurate results. Stereo vision systems can reconstruct the atmospheric information of a visual scene relatively accurately and have been widely used in mobile robot localization, navigation, obstacle avoidance, and map building. However, a major challenge in stereo vision systems is the problem of corresponding point matching, which significantly limits the application prospects of stereo vision in the field of robotics.
3. Multi-view vision system
Multi-view vision systems employ three or more cameras, with tri-view systems being the most common. They are primarily used to address the matching ambiguity problem in binocular stereo vision systems, thereby improving matching accuracy. The advantages of tri-view vision systems include fully utilizing information from the third camera, reducing erroneous matches, resolving the matching ambiguity inherent in binocular vision systems, and improving positioning accuracy. However, tri-view vision systems require more complex structural configurations due to the need for precise placement of the three cameras, and the matching algorithms are more intricate, consuming more time and exhibiting poorer real-time performance.
4. Panoramic Vision System
Multi-directional imaging systems with a large horizontal field of view have the outstanding advantage of a 360° field of view, which is unmatched by other conventional lenses. Panoramic vision systems can be achieved through image stitching or through catadioptric optical elements. Image stitching uses one or more cameras to rotate and scan the scene at a large angle, acquiring multiple consecutive frames of images in different directions, and then stitching them together to obtain a panoramic image.
