Pedestrian detection

[mvondracek]

Detect person in image from camera (#3). Try to use OpenPose (#4). Merge code with some form of an interface to the master.

• [x] single scene, single person, multiple cameras

EDIT: 1_extract_pose could be useful.

Input: images from multiple cameras Output: vertical line of detected person/people

[mvondracek]

@flaxh, what's the status, please? :)

[mvondracek]

• [x] testing data
  • [x] calibrated focal length, see triangulation.Camera.calibrate_focal_length
  • [x] coordinates and orientation of both cameras and coordinates of real subject, size of real subject, see test_triangulation.TestCameraDistanceTriangulation.setUp and /testing_data/
  • [x] single scene, single person, multiple cameras
  • [x] single scene, multiple people, multiple cameras
• [x] tests for our interface of OpenPose with our testing input data and expected single vertical line of detected person, EDIT: see openpose.test_povaPose (03688a18ed45f43c92a85683cc1f7d9ca793131f) and test_triangulation (90f3a01942f0dad372792b16db7e7cf189581b53).
• [x] connection from pedestrian detection to triangulation.Triangulation.locate, EDIT: see test_triangulation.TestCameraDistanceTriangulationSceneCorridor(TestCase) and test_triangulation.TestCameraDistanceTriangulationSceneLibrary(TestCase)
• [x] visualization, see #8

[mvondracek]

Output: vertical line ([1] - Nose xy, average( [2] - Right hip [3] - Left hip))

[mvondracek]

Class openpose.PovaPose.PovaPose is now connected to processing pipeline. Tests are in openpose.test_povaPose and test_triangulation. Communication with openpose.PovaPose.PovaPose is wrapped in detector.OpenPoseDetector(PeopleDetector). We are able to locate person in 6 m distance with delta 30 cm.

[mvondracek]

OpenPose is real-time multi-person detection library. However, our current interface takes 28 seconds to process two images. There should be a way how to speed it up. Could we for example process much smaller copies of input images?

@flaxh, could you please have a look at it? We definitely need the detection to be much faster.

callgraph with openpose.PovaPose.run_multi_person_detection

[mvondracek]

@flaxh, any updates?

[mvondracek]

@xstast24, please try to speed up person detection. subprocess.Popen could be useful for your 61f34405e96115116909090d6509d8304a686964 @flaxh could you please help with this?

[mvondracek]

Speed evaluation for repeated detection of person in testing image.

[mvondracek]

@xstast24 is working on integration of OpenPose binaries. 👍 #17