[WARN] [1671735223.499671649] [camera]: no pixel format selected, using default:
"BGR888"

is in the camera_node log file.

it gets this from my default set in config.txt

Autoload overlays for any recognized cameras or displays that are attached

to the CSI/DSI ports. Please note this is for libcamera support, not for

the legacy camera stack

camera_auto_detect=0
display_auto_detect=1
dtoverlay=vc4-kms-v3d,cma-512
dtoverlay=imx219
dtoverlay=imx219,cam0

Config settings specific to arm64

arm_64bit=1

from camera_ros.

There is no configuration file in this repo. By default, the node will select the first available pixel format. You can change that.

What is the encoding of the published mage? What happens when you change the format? How can I reproduce this?

from camera_ros.

I can reproduce this issue with the vivid driver on my desktop. After loading the kernel module via sudo modprobe vivid the new V4L2 device will show a test pattern with coloured vertical bars (note the red, blue, black bar order). You can check this also with VLC which lets you open a capture device:

To use this with libcamera, you have to use the vivid branch from the vivid fork https://git.libcamera.org/libcamera/vivid.git. When everything is set up correctly, you can use the device with the BGRA8888 format:

qcam -c vivid --stream pixelformat=BGRA8888

and it will show the test pattern correctly (red, blue, black):

When I then use this device with the ROS node and the BGR888 format:

ros2 run camera_ros camera_node --ros-args -p camera:=vivid -p format:=BGR888

the image will be published with the correct bgr8 encoding:

$ ros2 topic echo --once /camera/image_raw --field encoding
bgr8
---

but rqt will show the test pattern with switched red and blue colour bars:

from camera_ros.

from camera_ros.

This problem is not specific to the Raspberry Pi camera as I showed above with the vivid test driver.

When you mention the "config yaml file" do you mean the "Camera calibration file"? If you did not calibrate your camera yet, then this warning is normal. I just says that the camera calibration file cannot be found at the expected location because the camera was never calibrated.

from camera_ros.

from camera_ros.

Thanks for your reply. Is there a parameter to pass to camera_ros to flip the channels?

No, and there should be no such parameter because the colour format mapping between libcamera and the ROS image encodings should work correctly. I am going to look into this. In the meantime, you can simply change the encoding from bgr8 to rgb8 in order to have the channels switched.

from camera_ros.

from camera_ros.

I wrote a simple python script to visualise the channels of an image:

#!/usr/bin/env python3
import rclpy
from rclpy.node import Node
from sensor_msgs.msg import Image
import numpy as np
import cv2

class FormatVisualiser(Node):
    def __init__(self):
        super().__init__('format_visualiser')
        self.sub_img = self.create_subscription(Image, "/camera/image_raw", self.on_image, 1)

    def on_image(self, msg_img):
        dtype = np.dtype(msg_img.data.typecode)
        dtype = dtype.newbyteorder('>' if msg_img.is_bigendian else '<')
        img_raw = np.asarray(msg_img.data.tolist(), dtype=dtype)\
            .reshape(msg_img.height, msg_img.width, int(msg_img.step/msg_img.width))

        for i in range(img_raw.shape[2]):
            cv2.imshow(f"raw channel {i}", img_raw[...,i])

        if msg_img.encoding == "bgra8":
            img_rgb = cv2.cvtColor(img_raw, cv2.COLOR_BGRA2RGB)
            cv2.imshow("rgb", img_rgb)
            for i in range(img_rgb.shape[2]):
                cv2.imshow(f"rgb channel {i}", img_rgb[...,i])

            img_bgr = cv2.cvtColor(img_raw, cv2.COLOR_BGRA2BGR)
            cv2.imshow("bgr", img_bgr)
            for i in range(img_bgr.shape[2]):
                cv2.imshow(f"bgr channel {i}", img_bgr[...,i])

        cv2.waitKey(1)

if __name__ == "__main__":
    rclpy.init()
    exporter_node = FormatVisualiser()
    rclpy.spin(exporter_node)
    exporter_node.destroy_node()
    rclpy.shutdown()

When I use this with the BGRA8888 format (or BA24 in fourcc):

ros2 run camera_ros camera_node --ros-args -p camera:=vivid -p format:=BGRA8888

I can clearly see that the channels in memory are A, R, G, B , i.e. in inverse order (the alpha channel is supposed to be empty). This inverse order comes from the endianness (see https://en.wikipedia.org/wiki/RGBA_color_model#Representation).

Since the ROS message contains the endianness, I thought that the client is supposed to deal with the memory representation.

from camera_ros.

I (force) pushed an update to slect the target ROS encoding via the machine endianness. This should resolve your issue. Can you check?

from camera_ros.

Thank you, it works! Issue is resolved.

from camera_ros.

Nice. Then I will go ahead and close this issue.

from camera_ros.