adamvallance / celluview Goto Github PK

• Allow colours to be clustered and viewed on stream output.
• Add GUI element to adapt number of colour clusters.

PiCamera is deprecated in Bullseye 11. LibCamera2 would replace opencv videoio. This could potentially help with Flatfield correction, by allowing default processes to be turned off which may be causing colour issues.

Motor position labels should update upon startup by requesting from the sanga board.

Create diagrams, excluding UML, for readme

Looks like there is no/little red and the colors are messed up (when running on linux desktop through webcam).
Maybe an issue with BGR vs RGB?

Some versions of the OpenFlexure are liable to have jittery/shaky video streams on the microscopic level. Image stabilisation could correct this.

If a user or organisation has trained a convolutional neural network (CNN) on existing images and wishes to use this in real-time, they should be able to import the CNN and use it for real-time image processing.

• develop framework for CNN import
• add GUI element to enable/disable
• add GUI element for text output
• train small demonstration CNN for cellUview

User should be able to:

• see previous captures, including after power cycling
• update the default capture name with a text box
• click on a previous capture to expand (in new/switched window), see the settings, and restore these settings.

Functionality is quite slow on the pi when trying to do different image processing. Need to add threading to increase speed.

For the purposes of marking for realtime embedded processors 5, we need to refactor the code to match the course spec. by being well formatted and obeying SOLID principles, etc.

Perhaps allow for an option of saving the raw and enhanced frames? - use a copy constructor for the frame and enhance the copy

GUI window name currently set to MainWindow therefore showing up on the desktop as MainWindow. This should be replaced with cellUview

Create separate threads for the blocks so that the load can be balanced.

Was seeing a bug where after doing a flatfield capture and turning on, new flatfield captures were not possible.

Weird green centrally with pink at edges, can this be fixed? Maybe turning things off using libcamera, see #17, could help with this.

GUI should:

• Allow User to view stream of video frames as provided by the OpenFlexure
• Allow the user to vary parameters of image enhancement
• Allow the user to enable image analysis/classification
• Allow user to show hide separate control/gallery panels
• User can set note
• Change name of window from MainWindow to cellUview
• Gray out/hide motor controls if not plugged in

• After "bandpassing" by selecting a limited range of hues
• After increasing the saturation of selected hues

User should be able to change the default capture name (a member of the gallery class) from a textbox or other method (maybe a dialog box or static text box).

Increase saturation values for the selected hues to enhance colour. In HSV (hue, saturation, value) space.
If a user has stained cells with a given hue, it may be useful for them to increase the saturation of the hue of the stain.

After copying to /usr/bin logo doesn't show. Issue with the filepath?

When given a point-spread function of the optical setup, the video stream should be deconvolved to reduce blurring.

Segmentation faults have been observed when operating the motors branch. This must be fixed prior to merging.

Current frame size from the default opencv videoio is 640x480 - much smaller than raspberry pi camera. Maybe look at using libcamera to open and pass frames through existing framework.

std::list<std::pair<std::string, cv::Mat>> not being correctly assigned. This issue is not present on Raspberry Pi but should be investigated further.

Implement motor control software to interface with the Openflexure motor board. This would be required to implement a tiling procedure #14.

On occasion, exposure metadata will be encoded as value over 20000 upon capture rather than true value. Seems to happen randomly.

Issue in build.sh.

Implement existing logic for capturing an image, moving the stage, applying cross-correlation to determine the overlap and combining many images to build up a high-resolution image to be previewed by the user and exported. This requires motor control #13 to be implemented.

Following error message is printed when camera open fails:

'''
cellUview Error! blank frame grabbed
Please check your PiCamera microscope module is connected.<nterminate called after throwing an instance of 'std::system_error'
what(): Resource deadlock avoided
Aborted (core dumped)
'''

Would be ideal if program were to exit properly