GoWebScanCalProcessorNode Class Reference
Description
Represents an algorithm for processing node tiles containing a calibration bar, and generating node profile Z and Y offsets and vision Y offsets and gains. Refer to GoWebScanCal for a description of the system calibration outputs and usage.
GoWebScanCalProcessorNode executes a sequence of processing tasks for profile and vision data. If a task fails in the sequence, the processing does not continue and an error and issue type is recorded. Vision processing will not proceed if profile processing fails. Any warnings (issues that do not prevent processing from completing but may result in suboptimal calibration) that occur are recorded.
For profile processing, the sequence is as follows:
- The input profile data is inspected for any frame drops, time and distance incoherencies, and insufficient data.
- If user-defined obstructions are enabled, the profile data is masked (otherwise the chain region is detected automatically later).
- The input is reoriented in X (depending on X orientation) to standard system orientation.
- The input is resampled in Y using the resolution set in GoWebScanConfig_CalYRes().
- The input is reoriented (depending on Y orientation) and transposed so the leading Y edge is vertical.
- Z range outliers are removed using a histogram. Ranges from the chain should be removed in this step.
- Y edge row outliers (of the transposed data) are removed using a histogram. Removed data is filled (nearest neighbour).
- A line is fit to the filtered Y edges which forms the Y offsets. Depending on the travel orientation, the fit edges are first offset by the calibration bar width.
- A region of the Z filtered calibration bar data near the leading Y edge is used to determine the average Z range per spot. These ranges are filtered again using a histogram (filtered to keep only the 0.25 to 0.75 quantile). Removed data is filled (nearest neighbour).
- The Z offset for each spot is calculated as the filtered average Z value. For the top sensors, the Z values are subtracted from the calibration target height to form the offsets.
For vision processing, the sequence is as follows:
- The input vision data is inspected for any frame drops, time and distance incoherencies, and insufficient data.
- Each input image is reoriented in X (depending on X orientation) to standard system orientation.
- Each input image is demosaiced.
- Input images are resampled in Y using the Y resolution data from the GoGeoCal file, and a single image containing the calibration bar is formed.
- The input image is reoriented (depending on Y orientation) and transposed so the leading Y edge is vertical.
- Y edge row outliers (of the transposed data) are removed using a histogram.
- A line is fit to the filtered Y edges and the fit edges are sampled down to two points which forms the Y offsets. Depending on the travel orientation, the downsampled edges are offset by the calibration bar width.
- If GoWebScanSettings_CalDetectLocators() is enabled, a region near the center of the calibration bar image is determined, and each column is summed. The column sums are filtered via a histogram, and a center of gravity calculation is performed on the column sums to determine center of the hole. The X offset is calculated as the difference of the hole center (converted to an X distance) to the nominal X center.
- A raw input image approximately 1/4 from the leading Y edge is selected for gain calculations. The color channels are separated into different images.
- The individual channel image gradients are checked along X. If all rows have a large enough edge, this may imply the calibration bar does not span the full length, so a warning is produced.
- The channel images are vertically smoothened using a local histogram filter.
- The gain coefficients are calculated as the target intensity divided by the smoothened pixel value for each pixel per channel. The target intensity can be set by the user via GoWebScanSettings_CalUseVisionIntensity(), otherwise a default of 180 is used.
- The gain coefficients are downsampled to smaller size for convenient storage in the calibration file.
Inheritance diagram for GoWebScanCalProcessorNode:
The documentation for this class was generated from the following file: