Offset in Y direction

[Gabriele-97]

Thanks for this precious work!

I'm using this library to integrate the Azure Kinect in a robot system running in python.

I'm facing this issue: when extracting the 3D coordinates of the joints, there is a non-negiglible offset (up to 60cm) in the y coordinates of most joints (pelvis as well). X and Z positions are instead correct. Moreover, in the "combined image" the skeleton is perfectly aligned with the joints in any direction.
Has anyone noticed the same? Can you provide any reasonable explanation and suggestion for correction?
Thanks in advance!

The code I'm using is this one:

pykinect.initialize_libraries(track_body=True)
device_config = pykinect.default_configuration
device_config.color_resolution = pykinect.K4A_COLOR_RESOLUTION_OFF
device_config.depth_mode = pykinect.K4A_DEPTH_MODE_NFOV_UNBINNED

device = pykinect.start_device(config=device_config)
bodyTracker = pykinect.start_body_tracker()

cv2.namedWindow('Depth image with skeleton',cv2.WINDOW_NORMAL)
while True:
	capture = device.update()
	body_frame = bodyTracker.update()
	ret_depth, depth_color_image = capture.get_colored_depth_image()
	ret_color, body_image_color = body_frame.get_segmentation_image()

	if not ret_depth or not ret_color:
		continue

	for body_id in range(body_frame.get_num_bodies()):
		skeleton = body_frame.get_body(body_id).numpy()[:, :3]

		stampstring = str(time.time()) + " " +" ".join(str(r) for v in skeleton for r in v) + "\n"
		with open(filename,'a') as f:
			f.write(stampstring)

	combined_image = cv2.addWeighted(depth_color_image, 0.6, body_image_color, 0.4, 0)
	combined_image = body_frame.draw_bodies(combined_image)
	cv2.imshow('Depth image with skeleton',combined_image)

	if cv2.waitKey(1) == ord('q'):  
		break