Qbicz / real-time-shape-detection

CV/AI: use ARM Mali GPU on ODROID-XU4 single-board computer + OpenCL acceleration to achieve real time shape and orientation detection of fast moving objects

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Tune Canny threshold BEFORE TUNING SVM

Qbicz opened this issue · comments

@Qbicz could you please add some additional info?

Instead of tuning the threshold I will use fixed value 150 and try to perform convex hull to obtain closed contours

Good job, thank you! How many objects can we get from the images with the new algorithm?

I think it recognizes shape correctly in 90% of cases, when initial contour was not closed (and therefore will not be recognized).
I encourage you to run hu_moments_gahter.cpp when you find a spare moment.

After reopening:
changed sobel kernel size to 5 for better edges detection and adjusted thresholds.
Also added some useful comand line options.
Acorns not detectable will be fixed when the photo area willl be covered.