The idea of 3D vision is to obtain distance information from a scene. There are three widely used methods for this, each of which has its advantages and weaknesses. Stettbacher Signal Processing AG (SSP) has developed an inexpensive and powerful stereo camera based on its O-3000 camera technology.
It is so self-evident that we hardly notice it: Our two eyes not only provide us with an image of our surroundings. They also give us a three-dimensional impression of what surrounds us. This is particularly evident in the fact that we can very easily and accurately estimate the distance to an object in the field of view, which is one of the most important abilities in everyday life - apart from recognizing objects. Stereo cameras work according to the exact same pattern. Distance information can be obtained from the correlation of two images taken from different perspectives.
Unlike Structured Light or Time of Flight (ToF) methods, stereo vision is quite computationally intensive. It starts by first correcting for image distortion (due to the camera lenses) and very precisely equalizing the orientations of the two cameras. Both are typically done by an initial calibration and subsequent mathematical transformations of the images. Fortunately, modern and inexpensive embedded CPUs do this without difficulty. Subsequently, two major advantages over the alternative technologies become apparent: In contrast to Structured Light, no special illumination is required to operate the stereo camera, and compared to ToF, optical cameras provide high image resolution. This makes it possible to capture larger scenes and search for a (possibly moving) object in them, whereas ToF cameras usually must be aligned with the object. Depending on the task, SSP uses either a geometric method or machine learning with neural networks to search for an object in the image. Both methods complement each other perfectly and are extremely powerful, especially machine learning.
Stereo cameras evaluate two images of the same scene taken from slightly different perspectives. In the process, an object appears at different locations in both images. By correlating the images locally, both the distance to the object and the location in the room can be determined.
The picture shows a stereo camera from SSP. An IP67/69K housing is available upon request. The outer two cameras provide monochrome images. They are used to calculate depth information. Depending on the geometry of the setup, measurement accuracies down to the low per mil or even sub per mil range can be achieved. Optionally, the SSP stereo camera can be equipped with a third camera. This is mounted in the center and provides color images for a possible visualization of the scene. All cameras have a global shutter and are hardware synchronized.
SSP's stereo camera can be used in different ways: It can search for a single object in the image and measure it in space, or it can locate multiple objects of the same type and measure each individually, and finally, depth information can be determined for the entire image. Overall, the stereo vision system is an inexpensive, extremely flexible and very precise solution for object detection and localization.
© SSP. All rights reserved.
built by Andy