Best cameras for Jetson

Hi ,
Could you please let me know the best machine vision cameras we could use with Jetson .


1 Like

Hi vu3mmg,

The scope might be too big, maybe no the best, but only suitable, you should describe your project/use case more specific, then we can give the suggestion.
Besides, NVIDIA also enable Leopard imaging and Appro as Jetson platform partners to provide camera products, you could find their information from below link: [url][/url]


Thank you .
The application I am trying to build is object recognition in retail space .
I got a pointgrey Chameleon , with Tamron 80T4C with me .

Please let me know your expert opinion of re-using the above camera with Tx1 or TX2 .

Or is the built in camera good enough .

hello vu3mmg,

more details about your use case would be helpful.
is your use case need any specific sensor capability?
for example. how many megapixels? lens size? pixel size? output format?

In a very general sense here are my learnings on cameras with the TX1/2

USB Cameras:

  • Work fine in many use cases
  • Are easy to integrate
  • Can do a lot of the image work off-board (exposure control, frame rate, etc)
  • Many provide inputs/interrupts that can help time your application (e.g. interrupt on new frame)
  • Use CPU time due to USB bus, this will impact your application if it uses 100% CPU
  • Are not optimal for use of hardware vision pipeline (hardware encoders, etc)
  • Can work over long distances (up to max of USB standard)
  • Can support larger image sensors (1" and higher for better image quality and less noise)
  • Gives you high level control of the sensor/camera through a manufacturer's API
  • Often have tools/GUI around their API so that you can test out features without having to write code

CSI Bus Cameras:

  • Optimized in terms of CPU and memory usage for getting images processed and into memory
  • Can take full advantage of hardware vision pipeline
  • Can be a pain in the arse software/control wise since the libraries and support on TX platforms are not mature
  • Barebones features (exposure, etc done in TX1)
  • Short distances from TX1 only (10cm max usually) unless you use serialization systems (GMSL, FPD Link, COAXPress, Ambarella) which are immature and highly custom at the moment though some of the vendors bellow currently carry some models
  • Are mostly smaller sensors from phone camera modules but custom ones can be made at a price. The added noise from the smaller sensor can be mitigated a bit through the hardware denoise in TX1/2
  • Gives you access to low level control of the sensor/camera

In general, if your application is extremely intensive and you are fighting for every last drop of performance (high FPS, high resolution, etc) and you need low level control of the camera you want to be using a CSI camera, if your application is fairly light (i.e. you don’t need super high FPS, resolution etc) then USB cameras are fine, easier to use, and have better support from the manufacturers.

Lastly please note that there are other ways of getting images into the TX1 such as ethernet cameras (which I have not explored in practical situations), these also give you the flexibility of range from the TX1 and can be installed theoretically anywhere in a building and then networked to the TX1 using the buildings existing network infrastructure. Not as good for mobile applications though since you need PoE switches etc.

Some popular USB Camera manufacturers/suppliers:

  • Flir (was Point Grey Research) - 12+ week lead time on most models
  • Basler - 17+ week lead time on most models
  • iDS - Never used them but look nice
  • Ximea - Great cameras, more expensive
  • The Imaging Source - drivers/API are not as well documented as the others
  • Allied Vision - never used them, mostly different interfaces but have some USB models
  • Leopard Imaging - nVidia Partner, basic cameras though nowhere near features of above
  • There are many many more... Just google "Machine Vision Camera"
  • The other option is generic webcams which can be used directly by OpenCV without any custom drivers etc but give almost zero features and control on exposure, gain, framerate, etc

Some popular CSI manufacturers with TX1 support:

  • Leopard Imaging
  • e-Con Systems

Note that US/Canadian machine vision manufacturers have been hit by massive demand in the last 2 months. From what I have heard through the grapevine this is due to increased factory automation in the current political climate. This means that folks like Leopard and e-Con are some of the only suppliers you can get cameras from in reasonable time frames without having to import.

1 Like

Hi Jazza,
Thank you .

One more query . Can we develop software using TX2 and deploy it on TX1 .

Hi Jerry,

From your experience , could you please let me know the sensor size and pixels for a people tracking system.I am a novice in the area.The application is to track persons in a mall and capture the face .
One location where the camera has to be placed is near the check out counter , to capture the faces of customers and sales agents . I think the view angle has to be more or less straight or may be with in 30 degrees .


hello vu3mmg,

generally speaking, this depends on your algorithm complexity.
thanks Jazza to list CSI/USB camera features, please refer to Jazza’s list and consider your design purpose to determine your suitable camera sensor. thanks

I don’t know much about camera, will this help [url][/url]

Those are cameras used to take photographs with.

Computer vision typically use cameras with different interfaces, such as:

Especially the IP cameras are good for applications where the processing will happen in a different place than where the camera (or cameras) is/are located.

Thank you

SEE3CAM_CU30 (3.4MP with superior Lowlight performance), SEE3CAM_CU130 (13MP 4K ) are some of the popular USB 3.0 cameras from e-con Systems. It also comes with MIPI CSI-2 camera interface for TX1. Both these cameras are readily available for TX1 CSI-2 camera interface with driver support. It also supports Face detection and smile detection. Should be a good fit for your face capture application in a mall. Do contact for further discussions.

Jetson TX2/TX1 cameras:
e-CAM30_CUTX1 - 3.4 MP NVIDIA Jetson TX2/TX1 Camera Board (HD Camera)
13MP NVIDIA Jetson TX1 Camera Module - e-CAM130_CUTX1 (Ultra HD Camera - 4K)
e-CAM31_TX2 - 3.4 MP NVIDIA Jetson TX2/TX1 Camera Board (Autofocus Camera)
13MP NVIDIA Jetson TX2 Camera Board - e-CAM131_CUTX2 (13MP Camera)

@e-conSystems: do you provide a discount for the cameras listed?

@Andrey1984: Yes, You can avail the launch offer USD 50 off if you purchase before 3rd Aug 2017. Please visit our webstore - Buy e-con's USB Cameras | Modules | Boards | SOM


My questions is similar to vu3mmg, but a little different.

I am trying to build an application to perform object detection and facial identification. This is for a security system application, so I need a camera (or cameras) that can perform these actions over a variety of distances, short (1-10 yds), mid (10-30 yds) and long (30-100 yds).

After reading the above posts I reached out to e-Con systems and was recommended:
e-CAM132_TX2 - 4K Autofocus MIPI NVIDIA® Jetson TX2/TX1 Camera Board: [url]13MP 4K Autofocus MIPI NVIDIA® Jetson TX2 Camera Board
But I am concerned that since this sensor is so small it will not have the ability to do object detection and facial identification past shorter distances.

I looked through the great resource kayccc provided: Jetson Ecosystem | NVIDIA Developer
but since I’m a novice when it comes to cameras, I am having a hard time finding something that I’m confident fits my needs.

One camera that seems to fit what I’m looking for is the Bosch Dinion:
With P-iris lens:
But it doesn’t seem to be able to integrate into NVIDIA software.

I’ve reached out Basler AG also, but their cameras seem to only support fixed, static applications with a small focal window, which doesn’t fit my needs.

Can you please help me identify some cameras that may fit?



1 - 100 yards? Math and physics are against you.

If you have sufficient resolution at 5 feet (1.6 yards), you will need 20x the resolution in X AND 20x the resolution in Y to have the same resolution at 100 feet (30 yards) – that’s a 1.6 gigapixel camera if 4 megapixels is sufficient at 5 feet.

What these systems do in practice, is use either a single camera with adjustable zoom optics and adjustable orientation, OR use two cameras, one for wide field of view to identify areas of interest, and one zoomed-in camera with adjustable orientation to actually look at the area of interest.

You can use one detector / model to figure out “what things to focus on” based on the overview image, a simple servo control to point the zoomed camera at the location and perhaps adjust zoom/FOV, and then use a second model to do analysis of what the second camera sees.

I’d recommend prototyping this using USB cameras if possible, as those will be easier to source and test out for the Jetson.

Kindly look into below mentioned our smart camera video. It might be suit your requirement.

Read More: IP66 AI Smart Camera for Vision at the Intelligent Edge


Thank you for the Information provided above for the use of cameras with TX2.
I am currently using TX2 with 4 fisheye cameras using 2 USb 3.0 hubs connected, but the problem is I can make only two cameras work at a time.
If I give the command to run the 3rd camera, it says ‘No space Left on Device’

Please someone recommend me solutions how can I use all the 4 cameras together to work in realtime.

Can I connect the cameras using PCI-E to USb coverter?

Any suggestions will be helpful

what is the output of executing:

df -h
<b>nvidia@tegra-ubuntu</b>:~$ df -h
Filesystem      Size  Used Avail Use% Mounted on
/dev/root        28G   18G  8.2G  69% /
devtmpfs        7.7G     0  7.7G   0% /dev
tmpfs           7.7G   54M  7.7G   1% /dev/shm
tmpfs           7.7G   70M  7.7G   1% /run
tmpfs           5.0M  4.0K  5.0M   1% /run/lock
tmpfs           7.7G     0  7.7G   0% /sys/fs/cgroup
tmpfs           786M   56K  786M   1% /run/user/1001
/dev/mmcblk1p1   60G   33G   27G  56% /media/nvidia/9AE0-B49D

“No Space Left on Device” means “out of USB bandwidth” in this case, “df” won’t help at all in diagnosing that.

Are the cameras native USB 3 cameras? If they are USB 2, then they will run at 2 speed and use proportional time on the bus, which means even if you use a USB 3 port, you’re limited to how much data you could send on a USB 2 bus. The fix for that problem could be to replace the cameras with native USB 3 cameras. (These are hard to find, at least with low cost – I know the Stereolabs cameras are USB 3, and they can be used as “simple” cameras, although they don’t have “fisheye” optics.)

A PCI Express USB adapter card “should” work, assuming it has good Linux support. You may have to re-build the kernel to add support for that particular chipset (because while xHCI are all “standard,” they seem to also require workarounds at times?)