CODEWITHSUNDEEP
c++image-processingopencvobject-detection
enzom83
enzom83

Reputation: 8310

How to merge overlapped detections for each detected object?

I developed a software for testing object detection using HOG descriptor.

The presence of overlapping detections (multiple bounding box of different positions and sizes) for each object requires a procedure to combine them into a single bounding box that includes all the detections related to the same object. How to merge multiple detections for each object?

Upvotes: 2

Views: 1999

Answers (1)

Chris
Chris

Reputation: 8180

This can be accomplished using the OpenCV groupRectangles function, which clusters rectangles with similar sizes and locations according to a similarity parameter. Although, I believe this function may already be being used within the HOG detectMultiScale function according to the parameter group_threshold:

void gpu::HOGDescriptor::detectMultiScale(
    const GpuMat& img,
    vector<Rect>& found_locations,
    double hit_threshold=0,
    Size win_stride=Size(),
    Size padding=Size(),
    double scale0=1.05,
    int group_threshold=2
)

Of course you will have to try and tune the similarity criterion to suit your needs.

Upvotes: 2

Related Questions