OV9782 Pink hue vignette

silo

I'm using an Oak-D Pro Wide and I'm having an issue with the center color camera. Essentially there is a very strong pink hue around the outside edges of the image. Since I purchased the camera I have developing everything indoors and this issue never really occurred, or it was so minimal that I couldn't actually see it. However I started doing some testing outdoors last week and began noticing a pink hue that gets much stronger towards the edges of the image.

I've included a gif and a couple stills showing this. To be clear, those plants in the image are most certainly green and there is no purple/pink anywhere in the scene. I captured the gif/frames around 5:30PM while the sun was below the buildings/trees so that I could try to eliminate as much direct UV light as possible. But the pink hue is still quite strong.

Example

This particular issue is a real pain point for my application because I need a decent color image. For some background, the camera will be located on an autonomous platform and it needs to be able to recognize certain things in the environment that rely on color. My industry uses multi colored light poles as a means of communication between machines, people, and other machines. For example a combination of white, red, and orange signifies that the machine is powered on but not currently taking measurements so it is safe to approach. So not being able to discern between certain shades of colors when they are not in the central region of the image is a serious hinderance.

jakaskerl

silo
Pink hue is due to IR light passing through. Looks like your one does not have IR CUT filter (which it should have). You can RMA - https://www.luxonis.com/rma.

Apologies for the inconvenience.
Thanks,
Jaka

silo

Thanks for the response. I've seen an array of other posts with similar issues so I was starting to fear it was an actual hardware problem.

Examples

I am having one other issue that I was hoping to get your opinion on. I've attached two more images as examples. I'm noticing really poor image quality from the rgb camera. In the first image there is the closest light pole and some parking signs next to it. Those signs are only around 20-25 meters away, but they are extremely blurry. Also when you look at the closest light pole there is a significant amount of what almost looks like aliasing or stair stepping. However this pole is smooth all around. In the second image you can see how the sweet spot for the rgb camera is really only the central region and only within about 2-4 meters from the camera. Once you get outside of that zone everything turns into a blurry mess pretty quickly.

Do you know if the poor quality is somewhat normal for the OV9782 or if my unit has more issues than just the pink color? I read that the 9782 is generally going to have worse image quality than the IMX but this seems quite severe. I originally chose this configuration because it was one of the few that offered global shutters on the rgb camera as well. But if the global shutters have significantly degraded quality compared to the rolling shutters then I may have to rethink my approach.

jakaskerl

silo
What code are you using to capture those images? They seem very low res.

Thanks,
Jaka

silo

jakaskerl

This is the code I was using to save them. There is a lot of extra code in here but I wanted to give a reproducible example so it could be run.

def processDepthFromDevice(depth_cv2_frame, max_depth):
    # Handles/cleans the depth frame coming directly from the stereo node on the device

    depth = depth_cv2_frame.astype('float64')

    # Filter out undefined values
    depth[depth > max_depth] = np.nan  # Set everything past the depth cap to null/0
    depth[depth == 0] = np.nan  # Set everything with zero depth to nan

    # Create an image of depth for cv2
    depth_display_frame = (np.nan_to_num(depth) * 255 / max_depth).astype(np.uint8)
    depth_display_frame = cv2.applyColorMap(depth_display_frame, cv2.COLORMAP_HOT)

    return depth, depth_display_frame


def serialReader(shared, port, baudrate=9600, timeout=0.005):
    try:
        ser = serial.Serial(port, baudrate=baudrate, timeout=timeout)
        while True:
            if ser.in_waiting > 0:
                gps_timestamp = dai.Clock.now().total_seconds()
                gps_msg = ser.readline().decode('ascii', errors='ignore').strip()
                print(gps_msg.split(',')[0:7], end='\r')

                # Save the gps message if recording is active
                if shared.record_bool:
                    gps_msg_filename = str(int((gps_timestamp - shared.clock_shift) * 1000)) + '_gps.txt'
                    with open(os.path.join(shared.path_to_gps_msgs, gps_msg_filename), 'w') as gps_file:
                        gps_file.write(gps_msg)

    except Exception as e:
        print(f"[Reader] Error: {e}")

    finally:
        try:
            ser.close()
        except NameError:
            pass
        print("[Reader] Serial port closed")


# Main code
if __name__ == '__main__':
    # Set parameters for recording frames
    path_to_top_level_save = r''

    # Set up the parameters that need to be shared with the serial reader process
    manager = multiprocessing.Manager()
    shared_data = manager.Namespace()
    shared_data.record_bool = False  # This stays as false and gets changed by pressing 's' to record and 'p' to stop
    shared_data.clock_shift = 0

    # Set parameters for the oak-d pipeline
    fps = 30  # Sets the frames per second, also sets the min shutter speed as a consequence
    IMAGE_WIDTH, IMAGE_HEIGHT = 1280, 800
    DEPTH_CAP = 15000  # The max depth in mm that is allowed to be included


    # Create pipeline
    pipeline = dai.Pipeline()
    device = dai.Device()

    # Define the nodes
    camRgb = pipeline.create(dai.node.Camera)
    left = pipeline.create(dai.node.MonoCamera)
    right = pipeline.create(dai.node.MonoCamera)
    stereo = pipeline.create(dai.node.StereoDepth)
    imu = pipeline.create(dai.node.IMU)
    sync = pipeline.create(dai.node.Sync)
    x_out = pipeline.create(dai.node.XLinkOut)

    # Set stream names
    x_out.setStreamName("xout")

    # Set properties for rgb node
    rgbCamSocket = dai.CameraBoardSocket.CAM_A
    camRgb.setBoardSocket(rgbCamSocket)
    camRgb.setSize(1280, 800)
    camRgb.setFps(fps)
    camRgb.setMeshSource(dai.CameraProperties.WarpMeshSource.CALIBRATION)

    # For now, RGB needs fixed focus to properly align with depth.
    try:
        calibData = device.readCalibration2()
        lensPosition = calibData.getLensPosition(rgbCamSocket)
        if lensPosition:
            camRgb.initialControl.setManualFocus(lensPosition)
    except:
        raise

    # Set properties for left/right nodes
    # The disparity is computed at this resolution, then upscaled to RGB resolution
    monoResolution = dai.MonoCameraProperties.SensorResolution.THE_400_P
    # monoResolution = dai.MonoCameraProperties.SensorResolution.THE_800_P
    left.setResolution(monoResolution)
    left.setCamera("left")
    left.setFps(fps)
    right.setResolution(monoResolution)
    right.setCamera("right")
    right.setFps(fps)

    # Set properties for stereo node
    stereo.setDefaultProfilePreset(dai.node.StereoDepth.PresetMode.HIGH_DENSITY)
    stereo.initialConfig.setConfidenceThreshold(253)
    stereo.initialConfig.setMedianFilter(dai.MedianFilter.KERNEL_5x5)
    stereo.setLeftRightCheck(True)  # LR-check is required for depth alignment
    stereo.setSubpixel(True)  # This adds a lot of lag on 800p mono resolution
    stereo.setDepthAlign(rgbCamSocket)
    stereo.setDisparityToDepthUseSpecTranslation(False)
    stereo.setDepthAlignmentUseSpecTranslation(False)

    # Set properties for IMU node
    # imu.enableIMUSensor(dai.IMUSensor.ROTATION_VECTOR, 200)
    imu.enableIMUSensor([dai.IMUSensor.GAME_ROTATION_VECTOR, dai.IMUSensor.MAGNETOMETER_CALIBRATED], 100)
    imu.setBatchReportThreshold(1)
    imu.setMaxBatchReports(1)

    # Set properties for sync node
    sync_threshold = 20
    sync.setSyncThreshold(timedelta(milliseconds=sync_threshold))
    sync.setSyncAttempts(-1)  # Default, only sends message if all input nodes are in sync

    # Linking
    left.out.link(stereo.left)
    right.out.link(stereo.right)
    stereo.disparity.link(sync.inputs["disparity"])
    stereo.depth.link(sync.inputs["depth"])
    camRgb.video.link(sync.inputs["video"])
    imu.out.link(sync.inputs["imu"])
    sync.out.link(x_out.input)

    # Get max disparity to normalize the disparity images
    disparityMultiplier = 255.0 / stereo.initialConfig.getMaxDisparity()
    # print(stereo.initialConfig.getMaxDisparity())

    # Connect to device and start pipeline
    with device:
        # Set the debug level
        # device.setLogLevel(dai.LogLevel.INFO)
        # device.setLogOutputLevel(dai.LogLevel.INFO)

        # Get intrinsics
        calib_data = device.readCalibration()
        intrinsics_A = np.array(calib_data.getCameraIntrinsics(dai.CameraBoardSocket.CAM_A, dai.Size2f(IMAGE_WIDTH, IMAGE_HEIGHT)))  # This is correct one for aligned point cloud creation
        depth_hfov = calib_data.getFov(dai.CameraBoardSocket.CAM_A, useSpec=False)  # This shows the correct HFOV of 96.59 degrees when aligned, still need to figure out where to get VFOV

        # Start the serial reader process
        serial_process = multiprocessing.Process(target=serialReader, args=(shared_data, 'COM4'))
        serial_process.start()
        print('[Main] serialReader process started')

        # Set up the cv2 window
        cv2.namedWindow('rgb')
        cv2.namedWindow('depth')


        # Start the pipeline
        device.startPipeline(pipeline)

        # Turn on/off the IR dot projector or IR illumination
        device.setIrLaserDotProjectorIntensity(0.0)
        device.setIrFloodLightIntensity(0.0)

        # Start the queue
        queue = device.getOutputQueue(name='xout')
        queue.setMaxSize(1)
        queue.setBlocking(False)

        while True:
            # Get the message group from the camera
            msg_grp = queue.get()
            queue_timestamp = dai.Clock.now().total_seconds()
            imu_msg = msg_grp['imu']
            video_msg = msg_grp['video']
            disp_msg = msg_grp['disparity']
            depth_msg = msg_grp['depth']


            # IMU
            imu_roll, imu_pitch, imu_heading = calcOrientation(imu_msg.packets[-1].rotationVector)
            imu_quat = imu_msg.packets[-1].rotationVector
            imu_array = np.array([imu_quat.i, imu_quat.j, imu_quat.k, imu_quat.real])
            imu_timestamp = imu_quat.getTimestamp().total_seconds()
            # Get the magnetometer data
            imu_mag = imu_msg.packets[-1].magneticField
            imu_mag_array = np.array([imu_mag.x, imu_mag.y, imu_mag.z])

            # Print the orientation
            imuF = '{:.02f}'
            # print(f'Heading [deg]: {imuF.format(imu_heading)}  Roll [deg]: {imuF.format(imu_roll)}  Pitch [deg]: {imuF.format(imu_pitch)}', end='\r')

            # Get color frame
            rgb_frame_orig = video_msg.getCvFrame()
            rgb_frame_timestamp = video_msg.getTimestamp().total_seconds()

            # # Get disparity frame
            # disp_frame_orig = disp_msg.getCvFrame()
            # disp_frame = (disp_frame_orig * disparityMultiplier).astype(np.uint8)
            # disp_frame = cv2.applyColorMap(disp_frame, cv2.COLORMAP_JET)

            # Get the perpendicular depth frame directly from the device
            depth_frame_device_orig = depth_msg.getCvFrame()
            depth_frame_timestamp = depth_msg.getTimestamp().total_seconds()
            depth_device, depth_frame_device = processDepthFromDevice(depth_frame_device_orig, DEPTH_CAP)

            # Measure offset between Clock.now and device time in the first couple seconds after pipeline starts
            if imu_quat.getTimestampDevice().total_seconds() < 2.0:
                shared_data.clock_shift = abs(imu_timestamp - imu_quat.getTimestampDevice().total_seconds())

            # Save the messages if recording is active
            if shared_data.record_bool:
                # Save the rgb frame
                rgb_msg_filename = str(int((queue_timestamp - shared_data.clock_shift) * 1000)) + '_' + str(int((rgb_frame_timestamp - shared_data.clock_shift) * 1000)) + '_rgb.npy'
                np.save(os.path.join(path_to_rgb_msgs, rgb_msg_filename), rgb_frame_orig)

                # Save the rgb frame as png
                cv2.imwrite(os.path.join(path_to_rgb_png_msgs, rgb_msg_filename.replace('.npy', '.png')), rgb_frame_orig)

                # Save the depth frame
                depth_msg_filename = str(int((queue_timestamp - shared_data.clock_shift) * 1000)) + '_' + str(int((depth_frame_timestamp - shared_data.clock_shift) * 1000)) + '_depth.npy'
                np.save(os.path.join(path_to_depth_msgs, depth_msg_filename), depth_device)

                # Save the depth frame as png
                cv2.imwrite(os.path.join(path_to_depth_png_msgs, depth_msg_filename.replace('.npy', '.png')), depth_frame_device)

                # Save the IMU array
                imu_msg_filename = str(int((queue_timestamp - shared_data.clock_shift) * 1000)) + '_' + str(int((imu_timestamp - shared_data.clock_shift) * 1000)) + '_imu.csv'
                np.savetxt(os.path.join(path_to_imu_msgs, imu_msg_filename), imu_array, delimiter=',', fmt='%f')

                # Save the magnetometer array
                imu_mag_msg_filename = str(int((queue_timestamp - shared_data.clock_shift) * 1000)) + '_' + str(int((imu_timestamp - shared_data.clock_shift) * 1000)) + '_mag.csv'
                np.savetxt(os.path.join(path_to_mag_msgs, imu_mag_msg_filename), imu_mag_array, delimiter=',', fmt='%f')


            # Display the rgb frame
            cv2.imshow('rgb', rgb_frame_orig[::2, ::2])
            cv2.imshow('depth', depth_frame_device[::2, ::2])
            cv2.waitKey(1)



            # Check for key presses for recording functionality
            if keyboard.is_pressed('s'):
                if not shared_data.record_bool:
                    # Get the current time and date
                    current_datetime = datetime.now()
                    date_string = current_datetime.strftime('%m%d%y')
                    time_string = current_datetime.strftime('%H%M%S')

                    # Create the new directory to hold the frames for this recording session
                    os.chdir(path_to_top_level_save)
                    path_to_recording_session = os.path.join(path_to_top_level_save, str(date_string + '_' + time_string))
                    os.mkdir(path_to_recording_session)

                    # Write the intrinsic matrix
                    np.savetxt(os.path.join(path_to_recording_session, 'intrinsic.csv'), intrinsics_A, delimiter=',', fmt='%f')

                    # Make the folders for each message type
                    path_to_rgb_msgs = os.path.join(path_to_recording_session, 'rgb')
                    path_to_rgb_png_msgs = os.path.join(path_to_recording_session, 'rgb_png')
                    # path_to_disp_msgs = os.path.join(path_to_recording_session, 'disparity')
                    path_to_depth_msgs = os.path.join(path_to_recording_session, 'depth')
                    path_to_depth_png_msgs = os.path.join(path_to_recording_session, 'depth_png')
                    path_to_imu_msgs = os.path.join(path_to_recording_session, 'imu')
                    path_to_mag_msgs = os.path.join(path_to_recording_session, 'mag')
                    shared_data.path_to_gps_msgs = os.path.join(path_to_recording_session, 'gps')  # Shared with serialReader process
                    os.mkdir(path_to_rgb_msgs)
                    os.mkdir(path_to_rgb_png_msgs)
                    # os.mkdir(path_to_disp_msgs)
                    os.mkdir(path_to_depth_msgs)
                    os.mkdir(path_to_depth_png_msgs)
                    os.mkdir(path_to_imu_msgs)
                    os.mkdir(path_to_mag_msgs)
                    os.mkdir(shared_data.path_to_gps_msgs)

                    # Print to console
                    print('[Main] recording started\n')

                shared_data.record_bool = True
            elif keyboard.is_pressed('p'):
                if shared_data.record_bool:
                    shared_data.record_bool = False
                    print('[Main] recording stopped\n')

silo

jakaskerl

Is there anything in the code that might provide a clue as to why the images appear so low res despite the pixel dimensions?

jakaskerl

silo camRgb.setMeshSource(dai.CameraProperties.WarpMeshSource.CALIBRATION)

You are unwarping the image. If this might cause the blurriness. If you only try with rgb_video.py example (depthai-python/examples) do you still have the same issue?

Thanks,
Jaka