Can't find the fault in my code.

Hi,

I am doing a hand recognition project and I have a problem and I can't solve it.

That is the code our network was trained with. If I train the network it tells me that it has a accuracy of 99% percent but if I'm using the code below it dosen't recognize any hand gestures. It only shows me "search" which it only should do if its a black picture. Does anyone have an idea? Thank you!

clc;
close all;

%webcam
c = webcam('HD Pro Webcam C920');
load Network;

%handbox settings
Handbox=[0 0 300 300];

while true
    video = c.snapshot;
    
    %Video RGB to gray
    video_grey = rgb2gray(video);
    videoAdj = imadjust(video_grey);
    
    %Video RGB to BW
    BW = imbinarize(videoAdj);
    
    %BW inverte
    BW_inverse = imcomplement(BW);
    
    %Fill holes
    BW_IH = imfill(BW_inverse, 'holes');
    
    %change format [x x 1] --> [x x 3]
    BW_3D = repmat(BW_IH,[1 1 3]);
    
    %Create Handbox
    Handbox_new = insertObjectAnnotation(video,'rectangle',Handbox,'Handgeste');
    Handbox_crop = imcrop(BW_3D,Handbox);
    video_new = imresize(Handbox_crop, [227 227]);
    
    %classify
    label=classify(Network,video_new); 
    
    %show
    imshow(Handbox_new);
    title(char(label));
    drawnow;
end

[Walter Roberson]

That is a fairly small box compared to the field of view of the camera -- only about 1/3 by about 1/6 of the entire view (which is 1080 x 1920 for that particular camera). It is awkward to have to stand back far enough and put your hand all the way over to the edge of the camera view in order to get it inside that box.

[Image Analyst]

You might not have trained the network very well. Unfortunately we can't really tell that from Network, even if you did attach it, since it's basically a black box. We'd need to see all of your hundreds of training images. How many hand gestures were there? How many images did you use for each gesture?

[Marlene Patterer]

@Image Analyst

We took 500 images for each gesture, so for the numbers 0-5. Shell I send you one of each? Or what else may I give you so you have a better picture of it? Thank you!

@Walter Roberson

Which settings would you therefore use? Thank you!