The Cloud Motion Vector plugin uses the optical flow method to compute cloud motion vectors from the Sage sky camera images. The plugin utilizes the OpenCV calcOpticalFlowFarneback method to analyze the motion of clouds in images taken by the Sage sky camera. This information can be used to understand local weather conditions and predict short-term changes in solar irradiance. When combined with other plugins such as Cloud Cover and Weather Classification, the output of this plugin can be used to classify and track the movement of different types of clouds at various altitudes, which can be useful for meteorological studies [1].
The Farnebäck method is one approach to estimating the optical flow between two images. It assumes that the flow of pixels between two images is approximately constant within a local neighborhood [2]. The method calculates the optical flow by minimizing the sum of the squared differences between intensity values in the two images, subject to the constraint that the flow is approximately constant within a local neighborhood. The Farnebäck method is fast and simple, but it can be sensitive to noise and may not provide accurate results in the presence of large motion or changes in illumination.
The plugin uses the OpenCV calcOpticalFlowFarneback method to compute dense optical flow vectors from a pair of images taken at a user-specified time interval. The method returns an array of 2-channel floating-point images that has the same size as the input images, and each pixel of the output array stores the computed optical flow for the corresponding pixel of the input images. The optical flow is represented as a 2D vector, with the x and y components representing the flow along the x and y axes, respectively.
Currently, the plugin reads images from the sky-facing camera every 30 seconds and uses only the "Red" channel image. The mean direction and velocity of the cloud motion vectors are calculated and the results are published. The user can set the quality of the vector field, the time interval, and the RGB channels used for the computation.
Sensor needed: Sky Camera
Output: Direction and Velocity of mean cloud motion.
Input: Pair of images (30-60 sec interval)
Image resolution: Variable.
Inference time: 0.5 Sec
Model loading time: NA
The plugin has several arguments that can be set by the user:
--input: Path to input file or camera URL.--i: Time interval in seconds (Default: 30 sec).--c: RGB channels, 0=R, 1=G, 2=B (Default: 0, i.e., Red).--k: Keep fraction of the image after cropping (Default: 0.9).--q: Quality of the motion field (Default: 1). Sets averaging window, poly_n, and poly_sigma.1: Turbulent - Detailed motion field but noisy.2: Smooth - Lesser noise and faster computation.
--thr: Upload images if magnitude crosses this value (Default: 10).--segments: Number of segments for superpixel clustering (Default: 100).--seg_pub: Number of segments to publish in output (Default: 3).--oneshot: Process only one image pair and exit.
The plugin will publish the following data:
- Mean direction:
cmv.mean.dir - Mean velocity:
cmv.mean.vel
If the mean cloud speed exceeds a user-defined threshold (--thr), the plugin will also upload the processed images.
To query the output from the plugin, you can do so using the Python library sage_data_client:
import sage_data_client
# Query and load data into a pandas DataFrame
df = sage_data_client.query(
start="-1h",
filter={
"name": "cmv.mean.dir",
}
)
# Print results in DataFrame
print(df)
For more information, please see [Access and use data documentation](https://docs.sagecontinuum.org/docs/tutorials/accessing-data) and [sage_data_client](https://pypi.org/project/sage-data-client/).
### References
[1] Raut, B. A., Muradyan, P., Sankaran, R., Jackson, R., Park S., Shahkarami S. A., Dematties, D., Kim, Y., Swantek, J., Conrad, N., Gerlach, W., Shemyakin, S., Beckman, P., Ferrier, N., Collis, S., "Optimizing cloud motion estimation on the edge with phase correlation and optical flow". Atmospheric Measurement Techniques Discussions, 2023, 1-15. https://doi.org/10.5194/amt-2022-159
[2] Farnebäck, Gunnar. "Two-frame motion estimation based on polynomial expansion." _Scandinavian conference on Image analysis_. Springer, Berlin, Heidelberg, 2003.
### Credits
ecr-icon.jpg Wheat Field with Cypresses, Vincent van Gogh (Dutch), 1889.
https://www.metmuseum.org/art/collection/search/436535
https://www.metmuseum.org/about-the-met/policies-and-documents/image-resources
