Visualizing detections in the ROS code

[radu-diaconescu13]

Hello @Pandoro ,

I am trying to use the drspaam ros code and I am having some issues with visualization of the detections.

Namely, they look like this:

The purple circles representing detections seem to be misaligned.
The visualizations that are created during eval and training seem to work fine.

What drew my attention is this piece of code in the ros visualization method, i.e. detections_to_rviz_marker :
# circle r = 0.4 ang = np.linspace(0, 2 * np.pi, 20) xy_offsets = r * np.stack((np.cos(ang), np.sin(ang)), axis=1)

this is different to the way x,y coordinates are obtained in the train and eval code, i.e.
def rphi_to_xy(r, phi): return r * np.cos(phi), r * np.sin(phi)
with phi defined as:
def get_laser_phi(angle_inc=np.radians(0.5), num_pts=552): laser_fov = (num_pts - 1) * angle_inc return np.linspace(-laser_fov * 0.5, laser_fov * 0.5, num_pts)

From what I see the ros viz code is using a lower resolution of only 20 scan points and a hardcoded r value.
Do you think this could be the issue of the misaligned visualization?

Thank you

[Pandoro]

I think you are confusing two things. The circle you referenced is literally the red circle that is being plotted with a radius of 0.4 and 20 segments on that circle. The actual offset to the detection is computed as always and added to these red circles in the loop below in the same file:

2D_lidar_person_detection/dr_spaam_ros/src/dr_spaam_ros/dr_spaam_ros.py

Lines 121 to 138 in 99dd7a2

	# to msg
	for d_xy, d_cls in zip(dets_xy, dets_cls):
	for i in range(len(xy_offsets) - 1):
	# start point of a segment
	p0 = Point()
	p0.x = d_xy[0] + xy_offsets[i, 0]
	p0.y = d_xy[1] + xy_offsets[i, 1]
	p0.z = 0.0
	msg.points.append(p0)
	# end point
	p1 = Point()
	p1.x = d_xy[0] + xy_offsets[i + 1, 0]
	p1.y = d_xy[1] + xy_offsets[i + 1, 1]
	p1.z = 0.0
	msg.points.append(p1)
	return msg

As such it should not have anything to do with the actual alignment of the visualization and that has to come from elsewhere.