performance tests with MotionCam3D 4.1.3 and later

[bmegli]

Placeholder for results

If not specified:

• mode is Camera
• default texture source is Color
• OutputTopology is RegularGrid
• multipart transfer is used (not chunked)

Range only	Range only, TS = Laser
10.26	14.66

Range + 1932x1096 ColorCamera	Range + 1932x1096 ColorCamera, TS = Laser	Range + 1932x1096 ColorCamera, TS = Laser + MTU=9000
8.25	8.45	9.95

Range + 1288x730 ColorCamera	Range + 1288x730 ColorCamera, TS = Laser
9.86	9.9

Range + 3864x2192 ColorCamera, TS = Laser + MTU=9000
5.5

1680x1200 Scanner Range + 3864x2192 ColorCamera, TS = Laser + MTU=9000
2.2

Range + Intensity	Range + Intensity from Laser	Range + Intensity (chunked)
9.85	11.76	9.8

Device firmware notes

With MotionCam firmware 1.13.3 values I get are a bit lower, for example for

• Range + 1932x1096 ColorCamera, TS = Laser + MTU=9000
  • I get around 8.5 fps (vs 9.5)
  • similar to test without MTU change

The reason seems to be device not accepting any MTU changes above 1500 (GevSCPPacketSize)

After enabling jumbo frames and not modifying GevSCPPacketSize

arv-tool-0.8 control EnableJumboFrames=true

I get 9.7 fps which says that just changing that also changes MTU

[bmegli]

Range only

I had to enforce multipart data, otherwise it is sent as chunk buffer (not supported yet)

roslaunch camera_aravis photoneo_motioncam.launch enable_intensity:=false

rostopic hz /camera/Range/image_raw

Result

10 FPS even without texture

average rate: 10.257
	min: 0.068s max: 0.180s std dev: 0.01178s window: 306

[bmegli]

Range but enforce texture from Laser (disabled)

Texture is not streamed. This is just to test if it affects anything.

roslaunch camera_aravis photoneo_motioncam.launch enable_intensity:=false texture_source:=Laser

rostopic hz /camera/Range/image_raw

Result

average rate: 14.663
	min: 0.054s max: 0.129s std dev: 0.00522s window: 467

Comment

Ok, this means that device computes texture even if streaming of that component is disabled.

Ideally we would want to disable texture computation completely for added fps in Range

[bmegli]

Asked Photoneo about possibility to disable texture source completely or fastest setting in the absence:

• [GigEV] possibility to disable CameraTextureSource completely photoneo-3d/photoneo-cpp-examples#8

[bmegli]

This means that we should test Range + ColorCamera with texture source Laser

[bmegli]

Range + ColorCamera

roslaunch camera_aravis photoneo_motioncam.launch enable_intensity:=false enable_color_camera:=true

rostopic hz /camera/Range/image_raw

rostopic hz /camera/ColorCamera/image_raw

Result

average rate: 8.260
	min: 0.090s max: 0.145s std dev: 0.01162s window: 213

average rate: 8.252
	min: 0.080s max: 0.160s std dev: 0.01720s window: 207

[bmegli]

Range + 1932x1096 ColorCamera + enforce texture from Laser

Intensity not streamed, Laser texture is only to get more FPS

roslaunch camera_aravis photoneo_motioncam.launch enable_intensity:=false enable_color_camera:=true texture_source:=Laser

rostopic hz /camera/Range/image_raw

rostopic hz /camera/ColorCamera/image_raw

Result

average rate: 8.464
	min: 0.086s max: 0.235s std dev: 0.01461s window: 251

average rate: 8.449
	min: 0.072s max: 0.171s std dev: 0.01800s window: 177

Comment

Very small FPS gain.

Maybe chunked data would be better than multipart? (if sent before collecting everything

[bmegli]

Range + 1288x730 ColorCamera

roslaunch camera_aravis photoneo_motioncam.launch enable_intensity:=false enable_color_camera:=true color_resolution:=Resolution_1288x730

rostopic hz /camera/Range/image_raw

rostopic hz /camera/ColorCamera/image_raw

Result

average rate: 9.866
	min: 0.067s max: 0.167s std dev: 0.01278s window: 224

average rate: 9.862
	min: 0.068s max: 0.172s std dev: 0.01102s window: 227

Comment

Quite close to device computed color texture

[bmegli]

Range + 1288x730 ColorCamera, enforce Laser texture

Intensity not streamed, Laser texture is only to get more FPS

roslaunch camera_aravis photoneo_motioncam.launch enable_intensity:=false enable_color_camera:=true color_resolution:=Resolution_1288x730 texture_source:=Laser

rostopic hz /camera/Range/image_raw

rostopic hz /camera/ColorCamera/image_raw

Result

average rate: 9.899
	min: 0.056s max: 0.161s std dev: 0.01483s window: 278

average rate: 9.899
	min: 0.056s max: 0.161s std dev: 0.01483s window: 278

Comment

Unlike with Range only, texture source did very little difference

[bmegli]

Range + Intensity

The default

roslaunch camera_aravis photoneo_motioncam.launch

rostopic hz /camera/Range/image_raw

rostopic hz /camera/Intensity/image_raw

Result

average rate: 9.840
	min: 0.070s max: 0.159s std dev: 0.01294s window: 249

average rate: 9.863
	min: 0.064s max: 0.175s std dev: 0.01419s window: 237

Comment

Result like Range + 1288x730 ColorCamera but the other one should not need to align data

[bmegli]

Range + Intensity texture source Laser

roslaunch camera_aravis photoneo_motioncam.launch color_resolution:=Resolution_1288x730 texture_source:=Laser

rostopic hz /camera/Range/image_raw

rostopic hz /camera/Intensity/image_raw

Result

average rate: 11.758
	min: 0.066s max: 0.106s std dev: 0.00513s window: 210

average rate: 11.759
	min: 0.068s max: 0.103s std dev: 0.00476s window: 223

[bmegli]

Range + Intensity with chunked data (not multipart)

From:

• test performance with chunked data #22

roslaunch camera_aravis photoneo_motioncam.launch

rostopic hz /camera/Intensity/image_raw

Result

average rate: 9.805
	min: 0.069s max: 0.168s std dev: 0.01454s window: 222

We would not get anything in ROS Range as we are currently not testing for chunks attached to Image payload

Comment

This is exactly the same as multipart performance

[bmegli]

Range + 1932x1096 ColorCamera + enforce texture from Laser + jumbo frames OS MTU + jumbo frames aravis MTU

Setting MTU

sudo ifconfig enp60s0 mtu 9000

ip a
# ...
2: enp60s0: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 9000 qdisc mq state UP group default qlen 1000
# ...

Driver

roslaunch camera_aravis photoneo_motioncam.launch enable_intensity:=false enable_color_camera:=true texture_source:=Laser mtu:=9000

Test

rostopic hz /camera/Range/image_raw

rostopic hz /camera/ColorCamera/image_raw

Result

Before adapting MTU in camera_aravis, after stabilizing

average rate: 8.024
	min: 0.039s max: 0.503s std dev: 0.02526s window: 385

average rate: 7.952
	min: 0.023s max: 0.829s std dev: 0.03735s window: 554

After adapting MTU in camera_aravis with rosrun rqt_reconfigure rqt_reconfigure to 9000

average rate: 9.957
	min: 0.049s max: 0.192s std dev: 0.01319s window: 1127

average rate: 9.953
	min: 0.064s max: 0.183s std dev: 0.00905s window: 1385

Comment

MTU affects result drastically.

Especially in high bandwidth case like this.

MotionCam Firmware 1.13.3

Repeating experiment with firmware 1.13.3

• I can get only around 8.5 fps (v s nearly 9.5)
• enabling arv-tool-0.8 control EnableJumboFrames=true
  • I can get 9.7 fps