TopDownPeopleDetection

People silhouettes are detected as contours, with feature extraction like height at extremepoints/extremeties, height of center, etc. OpenCV based library, using depth images/ frames from a top down mounted depth camera, e.g. Kinect, Realsense etc.

This code has been developed by Katrin Hochschuh | Hochschuh&Donovan as part of the project Empathy Swarm before and as part of »The Intelligent Museum« at ZKM | Hertz-Lab.

Copyright (c) 2021 Katrin Hochschuh | Hochschuh&Donovan.

For information on usage and redistribution, and for a DISCLAIMER OF ALL WARRANTIES, see the file, "LICENSE.txt," in this distribution.

BSD Simplified License.

Description

In c++:

Create a cv::Mat of depth information, type CV_16UC1, from each frame: e.g. for Realsense Camera: cv::Mat depthMatrix = cv::Mat(cv::Size(w, h), CV_16UC1, (void*)depth.get_data(), cv::Mat::AUTO_STEP); depthMatrix.mul(depth_scale);

Before the frame loop create an instance of TopDownPeopleDetector: TopDownPeopleDetector peopleDetector;

Inside the frame loop call: peopleDetector.update(depthMatrix);

Afterwards you can retrieve the binary image that was created for debugging: cv::Mat testImage = peopleDetector.getOutImage();

and get a list of all detected BlobPersons: std::vector allBlobPersons = peopleDetector.getAllBlobPersons();

looping over this list, you can retrieve different information from the BlobPerson object and also display them by giving a reference to an image on which should be drawn, e.g.:

		//Contour center + radius
		cv::Point contourCenter = blobperson.getCenter();
		blobperson.displayMinEnclosingCenter(testImage);
		int contourRadius = blobperson.getRadius();
		blobperson.displayMinEnclosingCircle(testImage);

		// Point head center
		cv::Point3i headCenter = blobperson.getHeadCenter();
		blobperson.displayHeadCenter(testImage);
		blobperson.displayHeadHeightValue(testImage);

		// Extremepoints sorted by highest 
		std::vector<cv::Point3i> extremePointsSortedByHighest = blobperson.getExtremePointsByHeight();
		blobperson.displayExtremePoints(testImage);
		blobperson.displayExtremePointHeightsValues(testImage);

		//Convexhull
		std::vector<cv::Point> hull = blobperson.getHull();
		blobperson.displayConvexHull(testImage);

		//Contour simplified
		std::vector<cv::Point> contour = blobperson.getApproxCurve();
		blobperson.displaySimplifiedContour(testImage);

		//Display All Features of Detected BlobPerson//
		//blobperson.displayAll(testImage);

There is a helper function to display images that are in the wrong type: testImage = convertAnyMatTo_CV_8U_Mat(testImage); cv::imshow(window_name_depth, testImage);

Requirements

(Software) dependencies:

• OpenCV 4.1

Operating systems/platforms:

• Windows