Analysing larger images?

[afletch]

When using a pi camera v2 with a higher resolution, would it aid accuracy to be processing higher resolution images, and if so how is this achieved?

Changing CAMERA_WIDTH and CAMERA_HEIGHT seems to only zoom in, creating a larger output image but blurred/pixelated version with heavy cropping of all sides.

[pageauc]

Motion tracking uses captured frames from a camera video stream using the class PiVideoStream in speed-cam.py. The video stream resolution is based on the CAMERA_WIDTH and CAMERA_HEIGHT variables. This would be a portion of the Camera's full resolution frame .per http://picamera.readthedocs.io/en/release-1.12/fov.html Actual opencv motion processed is done with a cropped portion of the video stream based on

image_crop = image2[y_upper:y_lower,x_left:x_right]

at approx line 355 in function speed_camera() and this determines the detection accuracy so if you increase the CAMERA video stream resolution you might also need to increase the crop area variables if you wanted better tracking accuracy.

To get faster response a separate camera image is NOT taken. Since there is already a video stream frame, It is used for the final saved image and is just scaled up in size using WINDOW_BIGGER variable as a multiplier. Setting this to 0 or 1 saves camera video frame size image that is not resized. Using the original video frame avoids stopping/restarting video stream to setup and take a final image that takes time and may not capture the moving object due to the time delay in switching from stream to photo mode.

The camera video frame size can be made larger up to the maximum size video allowed by the particular camera module. This will have performance and memory consequences and possibly stability issues. V2 has better specs. This would slow frame processing speed but since OpenCV processing is based on a cropped area of the full size video frame, performance should be based on the image_crop variable above. The settings are a compromise since not all users will be running RPI3 with V2 camera module.

I have not tested very large video stream size settings too much. Note please be aware that the camera still internally crops full pixel capability per http://picamera.readthedocs.io/en/release-1.12/fov.html

I would suggest you experiment with various settings on your hardware to see how far you can push settings to maximize final image output and detection performance but please realize you will still not be able to get full camera pixel resolution per link above.

Your feedback is appreciated Regards Claude ....

On Sun, Feb 5, 2017 at 7:52 AM, afletch notifications@github.com wrote:

When using a pi camera v2 with a higher resolution, would it aid accuracy to be processing higher resolution images, and if so how is this achieved?

Changing CAMERA_WIDTH and CAMERA_HEIGHT seems to only zoom in, creating a larger output image but blurred/pixelated version with heavy cropping of all sides.

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/pageauc/rpi-speed-camera/issues/2, or mute the thread https://github.com/notifications/unsubscribe-auth/AFr1ZPFk57Pfu8pVEb-73vHqfHEaHdfRks5rZdQvgaJpZM4L3gJ1 .

-- See my YouTube Channel at http://www.youtube.com/user/pageaucp

[afletch]

Thanks for the detailed explanation. Despite having played with this extensively, it's still rather confusing what the values should be.

Would it be possible to provide some example settings for a higher-spec combination of Pi3 and cam2, using higher FPS and higher resolution tracking/images? I'm sure this would be useful for others too, and serve as an example from which to further test and tweak.

Thanks again.

[pageauc]

I just did a test on my RPI3 using CAMERA_WIDTH 1920 CAMERA_HEIGHT=1088 and WINDOW_BIGGER = 1

I did not change the webserver settings but I looked at the final 1080 output image and it is pretty good quality. I also did Not change the default opencv detection area settings. They are per below.

y_upper = 100 # Exclude event if y less that this value default=100 y_lower = 200 # Exclude event if y greater than this value default=200 x_left = 25 # Exclude event if x less than this px position Default=25 x_right = 295

NOTE the webserver size iframe would need to be increased as well but not be very usable on smaller monitor sizes.

FPS response was good, but remember opencv performance is base on the detection area restriction per x and y detection exclusion settings. If you make the detection area the full size of the image I would imagine performance would be slower but have not tested.

I suggest you do some testing.

So if you are trying to maximize final output image capture quality then you can just increase the camera width and height variables and set window bigger = 1 . This is only on a V2 camera module and suitable RPI with a quad core for best results.

Settings changes need to be optimized for hardware, performance, stability and results you are looking for.

Let me know how your testing goes and what results you are trying to achieve

Regards Claude .....

On Sun, Feb 5, 2017 at 9:15 AM, afletch notifications@github.com wrote:

Thanks for the detailed explanation. Despite having played with this extensively, it's still rather confusing what the values should be.

Would it be possible to provide some example settings for a higher-spec combination of Pi3 and cam2, using higher FPS and higher resolution tracking/images? I'm sure this would be useful for others too, and serve as an example from which to further test and tweak.

Thanks again.

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/pageauc/rpi-speed-camera/issues/2#issuecomment-277525975, or mute the thread https://github.com/notifications/unsubscribe-auth/AFr1ZDxsrl1NzmUtZPOuakxiCZwy_2V2ks5rZeeOgaJpZM4L3gJ1 .

-- See my YouTube Channel at http://www.youtube.com/user/pageaucp

[afletch]

Much appreciated, I shall take this and test further and feed back.

[pageauc]

fletch

I appreciate your feedback. I have updated speed-cam. makehtml.py now creates image links on image and data field to allow viewing full size images. I have also added config.py.240 and config.py.720 for lower power and higher power RPI's. The default is 720. You may need to customize the motion detection area but try not to make it too big since it will affect opencv processing speed. Also fixed a few potential issues in speed-cam.py if WINDOW_BIGGER and image_bigger variables are set to zero or less. This would cause a python crash. Note you will need to regenerate the html pages and will need to clear browser cache. I think you will like the changes.

To get the changes rerun the github curl command as the install script has changed

curl -L https://raw.github.com/pageauc/rpi-speed-camera/master/speed-install.sh | bash

Note when working with a larger image the various exclusion settings should be reviews such as min area Etc. since the pixel range id different. I have done some basic testing so if you notice something let me know via github issue.

Please let me know your testing feedback

Again thanks for your input. Claude ....

On Sun, Feb 5, 2017 at 11:10 AM, afletch notifications@github.com wrote:

Much appreciated, I shall take this and test further and feed back.

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/pageauc/rpi-speed-camera/issues/2#issuecomment-277533282, or mute the thread https://github.com/notifications/unsubscribe-auth/AFr1ZI2iPtI_bMlBsAfeCm9Po16N3uu4ks5rZgKggaJpZM4L3gJ1 .

-- See my YouTube Channel at http://www.youtube.com/user/pageaucp

[pageauc]

See latest release for updates

[pageauc]

I have updated speed-cam.py to ver 3.6 that now auto creates html path folder(s) and associated images symlink pointing to the image folder path. This is required for the webserver to work properly within its root html folder.

Please upgrade. I believe I have now covered most of the issues with the new release ...

Regards Claude ...

On Sun, Feb 5, 2017 at 2:50 PM, Claude Pageau pageauc@gmail.com wrote:

fletch

I appreciate your feedback. I have updated speed-cam. makehtml.py now creates image links on image and data field to allow viewing full size images. I have also added config.py.240 and config.py.720 for lower power and higher power RPI's. The default is 720. You may need to customize the motion detection area but try not to make it too big since it will affect opencv processing speed. Also fixed a few potential issues in speed-cam.py if WINDOW_BIGGER and image_bigger variables are set to zero or less. This would cause a python crash. Note you will need to regenerate the html pages and will need to clear browser cache. I think you will like the changes.

To get the changes rerun the github curl command as the install script has changed

curl -L https://raw.github.com/pageauc/rpi-speed-camera/master/speed-install.sh | bash

Note when working with a larger image the various exclusion settings should be reviews such as min area Etc. since the pixel range id different. I have done some basic testing so if you notice something let me know via github issue.

Please let me know your testing feedback

Again thanks for your input. Claude ....

On Sun, Feb 5, 2017 at 11:10 AM, afletch notifications@github.com wrote:

Much appreciated, I shall take this and test further and feed back.

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/pageauc/rpi-speed-camera/issues/2#issuecomment-277533282, or mute the thread https://github.com/notifications/unsubscribe-auth/AFr1ZI2iPtI_bMlBsAfeCm9Po16N3uu4ks5rZgKggaJpZM4L3gJ1 .

-- See my YouTube Channel at http://www.youtube.com/user/pageaucp

-- See my YouTube Channel at http://www.youtube.com/user/pageaucp