png from those raw frames, I used the tool, which additionally provides a few more format adjustment options. For raw videos, I have started to use the vooya tool, which sadly isn’t open source but provides an easy way for viewing and raw videos and extracting frames from it. At first, we activate the connection between the image signal processor and the image sensor. Most of the green nodes have one or more sinks (input) and at least one or more sources (output). The camera pipeline for the RkISP1 platform looks like this: Configuring the tool to stream RAW videos I found the vooya raw video player to be very helpful for that job as well as the tool as it provided a few more format options than vooya. After the recording of the video, I needed to find a way to view those videos and extract a frame. For both, I have to configure the camera pipeline in advance and I explain how I did it in the next section. So I used v4l2-ctl and yavta in order to capture raw frames. I am currently working a lot with the cam application from libcamera, but I noticed that it is currently not ready to capture raw video from the pipeline as the Raw stream role isn’t implemented yet. I found a few tools, that seemed to be viable for that job and a few that are currently not ready. My thought process behind this is that if the order changes then we should observe some weird colors as the brightness values of one color are used for a different color. I will capture every frame with the sensor format SRGGB10 and look at the result. In order to find proof that the Bayer order does change, I will try to capture RAW video from the camera pipeline and observe the result. So I assume that the bayer order should change, next I try to prove it by viewing the raw images with different flips and check for anomalies. When we now reverse the columns (vertical flip) we get: When I think about that sentence, my initial thought was like this: Additionally, it has a comment about flips, which states that the readout order of the sensor is reversed either horizontally or vertically depending on the flip variant. There is one data-sheet available online for this sensor, which mentions that the color filters are arranged in a Bayer pattern and that there is a line-alternating arrangement of the primary color BG/GR array. Why is it important to know if the Bayer order changes during a flip?īecause the order of the Bayer pattern is critical for the choice of the output Pixel-format. This small post is dedicated to documenting my journey of finding the answer to that question. I was asked if the Bayer order changes on those flips during the patch review and was not able to find a clear indication within the data-sheet. One of those new features is the ability to flip the image horizontally and/or vertically. If you want to know more about this you can take a look at this post. The image sensor was originally built by the Rockchip team and I try to upstream an improved version of it with a few extra features. I am currently working on porting and improving the driver for the OV13850 image sensor. An analysis of the RAW bayer images with the OV13850 image sensor on the RkISP1 platform Introduction
0 Comments
Leave a Reply. |