Skip to content

Latest commit

 

History

History
189 lines (145 loc) · 7.38 KB

README.md

File metadata and controls

189 lines (145 loc) · 7.38 KB

OpenNI2/NiTE2 Python Bindings

Python bindings for OpenNI2 and NiTE2.

Based on Primesense original bindings, with the bugs (well, at least some of them) fixed.

Example of NiTE2 usage:

import sys
from openni import openni2, nite2, utils

openni2.initialize()
nite2.initialize()

dev = openni2.Device.open_any()

try:
    userTracker = nite2.UserTracker(dev)
except utils.NiteError as ne:
    logger.error("Unable to start the NiTE human tracker. Check "
                 "the error messages in the console. Model data "
                 "(s.dat, h.dat...) might be inaccessible.")
    sys.exit(-1)

while True:

    frame = userTracker.read_frame()

    if frame.users:
        for user in frame.users:
            if user.is_new():
                print("New human detected! Calibrating...")
                userTracker.start_skeleton_tracking(user.id)
            elif user.skeleton.state == nite2.SkeletonState.NITE_SKELETON_TRACKED:
                head = user.skeleton.joints[nite2.JointType.NITE_JOINT_HEAD]

                confidence = head.positionConfidence
                print("Head: (x:%dmm, y:%dmm, z:%dmm), confidence: %.2f" % (
                                                                    head.position.x,
                                                                    head.position.y,
                                                                    head.position.z,
                                                                    confidence))

nite2.unload()
openni2.unload()

Another example of NiTE2. Display skeletons over the depth stream.

import sys
import argparse
from openni import openni2, nite2, utils
import numpy as np
import cv2

GRAY_COLOR = (64, 64, 64)
CAPTURE_SIZE_KINECT = (512, 424)
CAPTURE_SIZE_OTHERS = (640, 480)


def parse_arg():
    parser = argparse.ArgumentParser(description='Test OpenNI2 and NiTE2.')
    parser.add_argument('-w', '--window_width', type=int, default=1024,
                        help='Specify the window width.')
    return parser.parse_args()


def draw_limb(img, ut, j1, j2, col):
    (x1, y1) = ut.convert_joint_coordinates_to_depth(j1.position.x, j1.position.y, j1.position.z)
    (x2, y2) = ut.convert_joint_coordinates_to_depth(j2.position.x, j2.position.y, j2.position.z)

    if (0.4 < j1.positionConfidence and 0.4 < j2.positionConfidence):
        c = GRAY_COLOR if (j1.positionConfidence < 1.0 or j2.positionConfidence < 1.0) else col
        cv2.line(img, (int(x1), int(y1)), (int(x2), int(y2)), c, 1)

        c = GRAY_COLOR if (j1.positionConfidence < 1.0) else col
        cv2.circle(img, (int(x1), int(y1)), 2, c, -1)

        c = GRAY_COLOR if (j2.positionConfidence < 1.0) else col
        cv2.circle(img, (int(x2), int(y2)), 2, c, -1)


def draw_skeleton(img, ut, user, col):
    for idx1, idx2 in [(nite2.JointType.NITE_JOINT_HEAD, nite2.JointType.NITE_JOINT_NECK),
                       # upper body
                       (nite2.JointType.NITE_JOINT_NECK, nite2.JointType.NITE_JOINT_LEFT_SHOULDER),
                       (nite2.JointType.NITE_JOINT_LEFT_SHOULDER, nite2.JointType.NITE_JOINT_TORSO),
                       (nite2.JointType.NITE_JOINT_TORSO, nite2.JointType.NITE_JOINT_RIGHT_SHOULDER),
                       (nite2.JointType.NITE_JOINT_RIGHT_SHOULDER, nite2.JointType.NITE_JOINT_NECK),
                       # left hand
                       (nite2.JointType.NITE_JOINT_LEFT_HAND, nite2.JointType.NITE_JOINT_LEFT_ELBOW),
                       (nite2.JointType.NITE_JOINT_LEFT_ELBOW, nite2.JointType.NITE_JOINT_LEFT_SHOULDER),
                       # right hand
                       (nite2.JointType.NITE_JOINT_RIGHT_HAND, nite2.JointType.NITE_JOINT_RIGHT_ELBOW),
                       (nite2.JointType.NITE_JOINT_RIGHT_ELBOW, nite2.JointType.NITE_JOINT_RIGHT_SHOULDER),
                       # lower body
                       (nite2.JointType.NITE_JOINT_TORSO, nite2.JointType.NITE_JOINT_LEFT_HIP),
                       (nite2.JointType.NITE_JOINT_LEFT_HIP, nite2.JointType.NITE_JOINT_RIGHT_HIP),
                       (nite2.JointType.NITE_JOINT_RIGHT_HIP, nite2.JointType.NITE_JOINT_TORSO),
                       # left leg
                       (nite2.JointType.NITE_JOINT_LEFT_FOOT, nite2.JointType.NITE_JOINT_LEFT_KNEE),
                       (nite2.JointType.NITE_JOINT_LEFT_KNEE, nite2.JointType.NITE_JOINT_LEFT_HIP),
                       # right leg
                       (nite2.JointType.NITE_JOINT_RIGHT_FOOT, nite2.JointType.NITE_JOINT_RIGHT_KNEE),
                       (nite2.JointType.NITE_JOINT_RIGHT_KNEE, nite2.JointType.NITE_JOINT_RIGHT_HIP)]:
        draw_limb(img, ut, user.skeleton.joints[idx1], user.skeleton.joints[idx2], col)


# -------------------------------------------------------------
# main program from here
# -------------------------------------------------------------

def init_capture_device():

    openni2.initialize()
    nite2.initialize()
    return openni2.Device.open_any()


def close_capture_device():
    nite2.unload()
    openni2.unload()


def capture_skeleton():
    args = parse_arg()
    dev = init_capture_device()

    dev_name = dev.get_device_info().name.decode('UTF-8')
    print("Device Name: {}".format(dev_name))
    use_kinect = False
    if dev_name == 'Kinect':
        use_kinect = True
        print('using Kinect.')

    try:
        user_tracker = nite2.UserTracker(dev)
    except utils.NiteError:
        print("Unable to start the NiTE human tracker. Check "
              "the error messages in the console. Model data "
              "(s.dat, h.dat...) might be inaccessible.")
        sys.exit(-1)

    (img_w, img_h) = CAPTURE_SIZE_KINECT if use_kinect else CAPTURE_SIZE_OTHERS
    win_w = args.window_width
    win_h = int(img_h * win_w / img_w)

    while True:
        ut_frame = user_tracker.read_frame()

        depth_frame = ut_frame.get_depth_frame()
        depth_frame_data = depth_frame.get_buffer_as_uint16()
        img = np.ndarray((depth_frame.height, depth_frame.width), dtype=np.uint16,
                         buffer=depth_frame_data).astype(np.float32)
        if use_kinect:
            img = img[0:img_h, 0:img_w]

        (min_val, max_val, min_loc, max_loc) = cv2.minMaxLoc(img)
        if (min_val < max_val):
            img = (img - min_val) / (max_val - min_val)
        img = cv2.cvtColor(img, cv2.COLOR_GRAY2RGB)

        if ut_frame.users:
            for user in ut_frame.users:
                if user.is_new():
                    print("new human id:{} detected.".format(user.id))
                    user_tracker.start_skeleton_tracking(user.id)
                elif (user.state == nite2.UserState.NITE_USER_STATE_VISIBLE and
                      user.skeleton.state == nite2.SkeletonState.NITE_SKELETON_TRACKED):
                    draw_skeleton(img, user_tracker, user, (255, 0, 0))

        cv2.imshow("Depth", cv2.resize(img, (win_w, win_h)))
        if (cv2.waitKey(1) & 0xFF == ord('q')):
            break

    close_capture_device()


if __name__ == '__main__':
    capture_skeleton()

Captured screen image for above example.