Using libraries such as OpenCV, Keras and Tensorflow its possible to detect and recognize various objects.
Model used in application was trained on objects such as dark glass, white frame, strip, and clip.
Using the OpenCV library, the element edge, element position and element orientation are extracted.
Based on this, a portion of frame captured by cam is cropped and presented to neutral network for object recognition.
It is also possible to manually set the desired position and orientation of an item by clicking on it on the screen.
With the right configuration, edge detection can work in a variety of scenarios. As long as the object is detected correctly, the neutral network can predict a category to the detected object.
