Computer Vision News - March 2019

16 Computer Vision News 3D Stereo Vision Calibration Every month, Computer Vision News reviews a successful project. Our main purpose is to show how diverse image processing techniques contribute to solving technical challenges and real world constraints. This month we review a computer vision project by RSIP Vision : 3D Stereo Vision Calibration and Triangulation . Project Our client’s main goal in this project was to find the 3D location of certain points of interest. They had automated operations where they need to know the exact location of objects in space. The original system was not accurate enough for the client’s ambitions. For this purpose, we used two cameras to find the exact 3D location out of 2D image points. At the Proof Of Concept phase, the client defined some of the camera specifics, like the distance and the height of the objects of interest; the specs were of course expected to be different in the real production run. The two cameras system was needed to solve the depth part, transforming the 2D image into 3D points . Before starting the project, we did a simulation to verify whether additional cameras would enhance the performance in that environment: this was not the case, so for simplicity we decided to keep working with two cameras, optimizing the choice of angles, lenses and focus to be used in order to see the object with both cameras. There were two main challenges to obtaining accurate results in this project, that we solved using computer vision techniques. The first one was the stereo vision part and the triangulation part between the two 2D points: theoretically, this sounds like something that simple linear algebra equations can solve. In practice, these two points need to correspond: it has to be the same point in 3D that you see in both 2D cameras. In real life this is not always the case, due to distortions of the lenses , which we solved doing an appropriate calibration at the beginning of the project. This is a one-time process that fixes the distortion. We also needed to find the camera geometry, to do a correct triangulation. That means the camera’s intrinsic parameters as well as its rotation and translation to get to the coordinate system that we want and the relation between the cameras. Of course, in this process of finding the parameters you also have some errors and you need to be very organized and punctual in all the steps of the calibration, which involves repeating some of the parts to get better results. In this case, we used an object with a calibration pattern in the form of a known geometry: a chessboard was chosen in this and in many other cases since it has a grid with boxes of the same size and a repeated pattern, with straight lines, so that you can fix the distortion (by finding the distortion coefficients, so that all the corners appearing on straight lines in reality, will appear on straight lines on the image as well). By taking many images of the chessboard in different locations and using optimization algorithms, we were able to fine-tune the parameters and find the best ones to satisfy these conditions, solving the calibration problem. The need to cover most of the field of view with a single chessboard was challenging, since the field of view was pretty large and the chessboard limited in size. We solved this challenge by Aliza Minkov