How can I define a region of interest for object detection using a cascade detector?

I have trained a cascade detector to detect vehicles in a video, having problem with defining a ROI for the detection process.

detector = vision.CascadeObjectDetector('04052014_far_car4_good.xml', 'MinSize', [32 35], 'MaxSize', [300 300], 'ScaleFactor', 1.1, 'MergeThreshold', 4);

img = imread('00003_221.png');

bbox = step(detector, img); detectedImg = insertObjectAnnotation(img, 'Rectangle', bbox, 'Car'); figure; imshow(detectedImg);

I don't want to use a binary mask, I want to reduce the detection cost by defining a ROI. Suggestions?

Thanks a lot, Yoad.

Answer 1

[Dima Lisin]

vision.CascadeObjectDetector does not support specifying an roi directly. But you can always crop the image, e.g. using imcrop, and pass the cropped image into step().