MATLAB: How to detect multiple objects in a video file and label these objects with bounding boxes.

Computer Vision Toolboxconvolution neural networksdeep learningDeep Learning ToolboxImage Processing Toolbox

Hi, I want to know how I can create my own object detection using existing convolution neural networks like alexnet and have this applied to a video file that shows the bounding boxes of those objects.

Where are good places and sources so I can understand how to do this?

Best Answer

Watch Joe Hicklin's demo video, that does exactly that: MathWorks Deep Learning in 11 lines of code

Related Solutions

MATLAB: Mouse face detection in video

You'd probably have to use deep learning for that. Use transfer learning CNN. There should be a demo for that in the Deep Learning Toolbox.

MATLAB: Problem with bounding boxes.

You're not passing in a bounding box. boxes(idx,:) is a bounding box for idx = 1, 5, 9, etc. (every 4) but not for 2,3,4,6,7,8, etc. Just look at what boxes is. It's [x1,y1,w1,h1,x2,y2,w2,h2,x3,y3,w3,h3,.....] So when idx = 3, say, boxes(idx,:) = w1. Since boxes is a 1-D array, the ,: does nothing - it's just looking at the idx value. You'd have to skip by 4 and take just 3 of them:

for idx = 1 : 4 : m
    thisBoundingBox = boxes(idx:idx+3);
    if (bboxOverlapRatio(thisBoundingBox, bbox) > 0)

or better yet , don't concatenate the boxes and loop over them at all , but just loop over the regions:

for r = 1 : length(lframelabel)
    thisBoundingBox = lframelabel(r).BoundingBox;
    if (bboxOverlapRatio(thisBoundingBox, bbox) > 0)

Best Answer

Related Solutions

MATLAB: Mouse face detection in video

MATLAB: Problem with bounding boxes.

Related Question