Linking detections of the multiple network based on the pipeline

Hi Team,

I am trying to build an application where I detect the face of the person and then track his activities through post estimations. It is done in a single pipeline. I get the output in the Python callback function. The output I get is separate detection of face and person. But I am unable to find which face detection is of which person. Is there a way to link the face detection object to the person detection object? Please help me out.

Hi @ap20feb,

One approach you could take is to verify whether the bounding box (bbox) from the face detection falls within the bounding box of the person detection. You can do this by comparing the xmin, ymin, xmax, and ymax coordinates. If the face bbox is inside the person’s bbox, you can link the two detections.

Yes, this could be one of the possible solutions but there would be edge cases where this will fail. Luckily, in my case, pose estimation also gives left & right eye keypoints. if these keypoints are in face BBox then we can say that the face belongs to that particular person.

But if we could have a pipeline ID, then it would have been easy to identify and link detections

@ap20feb Identifying frames is not a problem. The challenge lies in associating faces with specific individuals when multiple people are detected in the same frame.