I am a beginner of haar detection and i have trained a haar descriptor with 900 images of a bottle obtained from the extraction of the frame from the video of the bottle.
i use cascade trainer gui.
by my problem is that i just succeed of detecting parts of the bottle and drawing the associated rectangles.
can you help me to understand the problem : have i sufficient positive images? is my parameters right? have i to make some treatments of the video/images? have i to merge some samples?
furthermore, can i calculate the size and position of the whole original image from a part detection?
