Lot of stereo cameras use random pattern projector to improve surface details required to stereo matching. Do you have any experience with that? What devices are market available?
Hi @macieksz
pattern projectors cannot be used with ZED cameras because they usually are based on IR light and all the ZED cameras are equipped with IR filters for daylight.
Usually, stereo cameras using pattern projectors have two greyscale sensors without IR filters to acquire IR patterns and extract depth information, plus a third RGB sensor to acquire color information.
@Myzhar This is an excerpt from Middlebury Stereo Dataset, pdf here -
We use ViewSonic 1024×768 DLP projectors, thus 10 patterns are sufficient in each of u and v. Instead of standard Gray codes [29] we use maximum min-stripe-width Gray codes [9] for increased robustness to interreflections and defocus.
Similarly, intel realsense D400 series (stereo cameras) can also be used with projector - Projectors for D400 Series Depth Cameras.
Can you cross check and provide your feedback?
Hi @saurabheights
Welcome to the Stereolabs community
Intel RealSense is an ACTIVE stereo camera, it uses two monochrome low-resolution CMOS sensors without IR filters to acquire images couples and obtain depth maps enhanced by artificial visual features generated by IR projectors, and it has a third hi-res RGB CMOS sensor to add color information.
ZED cameras are PASSIVE stereo cameras. We use ONLY visible light information to extract depth information without the needing of adding artificial information by using an IR projector.
ZED can work natively in sunlight conditions and the lenses have IR filters to remove the infrared frequencies of the visible light that would generate reddish images.
For this reason, you cannot take advantage of an IR projector with a ZED camera.
@Myzhar Please check the links I shared above related to Realsense as well as Middlebury dataset. In the provided link, it is a custom use case where realsense is utilizing consumer grade projector (not IR, but Image projector) to enhance depth capture of textureless regions. Middlebury Stereo dataset on the other hand uses coded texture maps to capture high quality depth maps using ViewSonic 1024×768 DLP projectors. Note - ViewSonic are customer grade projectors, not IR Projectors.
P.S. The reason I wanted to try this approach as we are using Zed 2 camera to capture a 3D Volume however, Zed 2 results are quite worse when the region is almost with no texture. Hence, I thought projecting a texture image as provided by Intel Realsense - https://files.readme.io/c68a821-fig_6.png would be helpful for Zed as well.
If you use visible light you can do that. Please share your results.