objectDetection (Report for single object detection) and YOLO.

Hello everyone!

I need some help with using the Object Detection function in YOLO to obtain detection results based on frames. I've tried various codes, but the results are not what I expect. Any advice or guidance would be greatly appreciated.

for example

Video = VideoReader("tr.avi",CurrentTime=0);

player = vision.VideoPlayer;

detections = {};

measurementNoise = 10;

while hasFrame(Video)

frame = readFrame(Video);

frameIndex = round(Video.CurrentTime * Video.FrameRate);

[bbox, score, label] = detect(detector, frame);

if length(detections) < frameIndex

detections{frameIndex} = {};

end

for i = 1:length(score)

detections{frameIndex}{end+1} = objectDetection(frameIndex, bbox(i,:), ...

"MeasurementNoise", measurementNoise, "ObjectAttributes", struct("Score", score(i)));

end

Object = insertObjectAnnotation(frame, "rectangle", bbox, label, "Color", "green");

step(player, Object);

end

Thank you!

Answer 1

[Samay Sagar]

Hi Bashir,

When working with YOLO for object detection on video frames, it is important to ensure that your setup is correctly configured and optimized for the task at hand.

Ensure that your YOLO detector is properly loaded and compatible with your MATLAB version. If you have a pretrained detector saved in a MAT file, you can load it using:

pretrained = load('yolov2ResNet50VehicleExample_19b.mat');
detector = pretrained.detector;

If you do not have a pretrained network, you can train your own model by referring to the following documentations:

https://www.mathworks.com/help/releases/R2024a/vision/ug/train-an-object-detector-using-you-only-look-once.html#ObjectDetectionUsingYOLOV2DeepLearningExample-6

To create your own YOLO detection network, refer the following documentation:

https://www.mathworks.com/help/releases/R2024a/vision/ug/getting-started-with-yolo-v2.html

Here is how you can use object detection with a pretrained network for vehicle detection:

Video = VideoReader("sampleVideo_640x480.mp4");
player = vision.VideoPlayer;
pretrained = load('yolov2ResNet50VehicleExample_19b.mat');
detector = pretrained.detector;
detections = {};                
measurementNoise = 10;

while hasFrame(Video)
    frame = readFrame(Video);
    frameIndex = round(Video.CurrentTime * Video.FrameRate);
    
    % Perform detection on the frame
    [bbox, score, label] = detect(detector, frame);
    
    % Store detections
    if length(detections) < frameIndex
        detections{frameIndex} = {};
    end
    for i = 1:length(score)
        detections{frameIndex}{end+1} = objectDetection(frameIndex, bbox(i,:), ...
            "MeasurementNoise", measurementNoise, "ObjectAttributes", struct("Score", score(i)));
    end
    
    % Annotate and display frame
    Object = insertObjectAnnotation(frame, "rectangle", bbox, cellstr(label), "Color", "green");
    step(player, Object);
end