1 Topic by leroyoddie

  • leroyoddie
  • Member
  • Offline

I would like to setup a dual high speed camera system for Kinovea. The application is clinical gait analysis with video synchronized from the sagittal and frontal planes. It seems one of the newer Kinovea releases significantly improved workflow and I want to make sure my hardware selection takes advantage of those improvements.

The order of priority is the fastest workflow, then maximum resolution, then highest frame rate. The goal for the fastest workflow is to minimize administrative tasks such as file transfers, synchronizing video frames, mouseclicks, etc. However, it is not evident how hardware choices may impact workflow. Being new to Kinovea, this leads to multiple questions regarding hardware possibilities and optimization:

1. A high frame rate is desired, but maximal resolution takes priority. A frame rate between 400-1000fps is desired. There are some Android phone cameras that can do 1080/1000fps. However, it is not clear if a setup with smartphones is efficient. Can a networked dual smartphone be as efficient as using machine vision cameras?  I'm referring to the time/effort it takes to setup/record/process the videos.

2. At higher frame rates of human motion (walking/running), is a rolling shutter a concern?  Or should one only consider a global shutter for this application?

3. Assuming a global shutter is used, can any Daheng machine vision camera be used? It is not clear if only certain models are supported.

4. If a global shutter is recommended, it seems the Daheng MARS-900-120GTC is the best option for my application (highest frame rate and resolution). However, it is a 10GigE camera, not USB 3. Does Kinovea support 10GigE cameras? If so, are there pros/cons to using 10GigE vs. USB cameras?

5. If 10GigE cameras are supported, are they any addition hardware requirements than a two 10GigE ports (e.g, dual port 10GigE capture cards)?

6. Assuming two Daheng MARS-900-120GTC cameras can be used, what would be the CPU, GPU, and RAM requirements?

I apologize for all the questions, but I could not find answers after studying Kinovea documentation and several hours searching the forum and Internet. It seems most of the applications discussed are for minimizing hardware cost, which is not my goal (although budget is not unlimited). Hopefully any feedback here will also be beneficial to others hoping to setup similar systems in the future.

2 Reply by joan

  • joan
  • joan
  • Admin
  • Offline

Hopefully someone with more hands on experience can come along.

A smartphone setup is different. To capture high speed footage with a smartphone you must use the smartphone itself for capture. When the smartphone does high speed it records to its onboard memory and then transfer it to its own storage. Then you have to transfer from that storage to a file on your PC. If you can somehow have access to the phone storage from the PC then it can be a time effective option. Capturing a smartphone camera feed from Kinovea capture module is not going to deliver the same performance as machine vision cameras.

Maybe someone else has a different take on this but in my opinion rolling shutter in human activities is mainly a concern for the objects we use, implements, balls, etc. because they can move much faster than the human body itself. Body segment movement under the athlete own force isn't very fast in the grand scheme of things. This is also true for frame rate in my opinion, you don't really get more information from a 2000 fps than a 1000 fps one when looking at body segments, we can't produce the kind of acceleration that would make a difference. (Not true for the objects we throw or the tip of things we hold though). Rolling shutter is directional, if you know the maximum extent of motion expected you can orient the camera to minimize it.

3 Reply by leroyoddie

  • leroyoddie
  • Member
  • Offline

Thank-you for the insight Joan!  It was my sense that the smartphone approach would not be as efficient as other options such as machine vision cameras, so I won't consider smartphones further.

Given your take on machine vision perhaps not be necessary for human movement (in this case walking/running), that opens the possibilities to other camera options which might have better resolution. This leads to the question if other camera types can be just as fast from a workflow perspective (time from capture to viewing) as machine vision camera?

It seems machine vision cameras may still be the best solution for my application. The ability to use a raw stream format to limit hardware requirements is may reduce hardware costs with minimal increase in workflow time.

4 Reply by Reiner

  • Reiner
  • Member
  • Offline

To answer your question some more information is needed.
What is your intention to use a very high resolution?
It would make things much easier to use a lower resolution around 800x600 or 1000 x 800, having the amount of data to be transferred in mind. Using this resolutions, a much cheaper camera can be used.
If you go up to the resolution of your selected camera (4192X2160), you will have significant problems to use a high frame rate due to the limitation in data transfer. Therefore, Getimage recommends a frame grabber that will create higher costs ( 2x500 Euro). if you select this solution, you have to check if you can use the high frame rate AND the high resolution at the same time. However, the file size of the created video will be huge.

In any case, do not use a rolling shutter camera. You will have a lot of image deformation (bending or extrusion of the objects).

All Daheng cameras that use the Galaxy driver can be used with Kinovea (as far as I know).

All GigE cameras have a lower data transmission limit compared to USB3.0. Maybe using a frame grabber can enhance it, but you should check what the limit is.
The limitation of cable length of USB can be overcome to a given amount. Using high quality cables you might extend the system up to 10m. Using special cable also to a higher distance (see Basler webpage for examples).

Beside frame rate the shutter speed is a very important factor and most often the more critical value. I suggest that you would have to go down to about 500-1000 microseconds to get a crisp image in your video. Otherwise you will have a lot of blurr in it if you look on the video image by image.


Regarding frame rate. In normal gait (about 2-5km/h) I would estimate the speed of the moving leg at about 20km/h. That is 5,5m/second. Using 100fps, you get an image every 5cm of movement, going up to 500 fps every 1,1cm.

Do you use your application indoor? If so, be aware that you may need a lot of light if your shutter speed goes down to 500-1000 microseconds. An absolute flicker free lighting is necessary.

The much cheaper cameras of Daheng already will give you a lot of options to adjust all the necessary parameters.

Two recommendations:
MER2-160-227U3M
1/2.9“ sensor IMX273, resolution 1440x1080, 227 fps, USB3.0

MER2-230-168U3M
1/1.2“ sensor IMX174, resolution 1920x1200, 168 fps, USB3.0

Similar cameras from IDS and Basler are available, but at a higher price.

I recommend a mono camera due to better light intensity and 1/3 of the file size compared to a color camera.
I use 3 of the 168U3M at the same time in video golf recording. The image quality if pretty good. I also have a camera from Basler with the IMX273 that gives pretty good images.

Keep in mind that you can increase the frame rate of both cameras by reducing the image height (not width). Reducing it for example to 1/2 of the specification, the fps rate increases by a factor of 2. This works proportionally.

The 1/1.2“ sensor camera has a better image than the 1/2.9“ and needs less amount of light. However, the lenses for the 1/1.2“ camera are much more expensive.
The camera plus lens cost of the 1/2.9“ camera is around 500 Euro, of the 1/1.2“ camera around 1100 Euro.

Selecting the right lens for your application may be a further topic. But first you should carefully define your geometrical situation of distance to the object and  the field of view you would like to record. On the Basler website you may find a valuable tool to calculate the FOV for a given image sensor/lens combination.

Regarding the workflow you might look on the Golfsimulator-forum. A versatile script, not only applicable for Golf, is available. It can load a complete setup with one click once it is defined and do a lot of post processing to extract videos, single or stacked images, user management etc..
Search for: „Comprehensive programs supporting video golf recording using Kinovea“

5 Reply by leroyoddie

  • leroyoddie
  • Member
  • Offline

Thank-you for the detail response Reiner!  I was hoping you would chime in as I have learned a lot from your other forum posts. I have been sick lately and am just getting caught up.

The intended use is more acurate data (higher resolution & frame rate) for clinical investigation (angles & linear translation). Video analysis will always be indoors with supplementary flicker free lighting and most likely always on a treadmill. There will be plenty of space around the treadmill so the FOV is not critical, although longer lenses will help reduce parallax. There will not be a need for long USB cables as the computer will be located near the treadmill. Thus, the camera data transfer cables should be <2-3 meters each.

FWIW, it is difficult to correlate walking velocity to limb velocity as the former is relatively linear and the latter is angular. My guestimate for direct comparison is the swing phase advancing limb is approximately 3x faster than walking velocity, accounting for acceleration and deceleration.  However, my interest is only in stance phase walking in which the velocities are closer to walking velocity - much slower than the swing phase limb.

In setting up a system, there are too extremes - as low cost as possible (golf setup at home) and no cost limits (big corporation, hospital, university, etc.).  My hope is to find the setup that is in between the two extremes - a value based approach, yet closer to the budget end of the spectrum. For example, if doubling the resolution costs 2x for the equipment, there is value in the higher resolution. Conversely, if double the resolution costs 10x as much, then it is not worth it to me.  It seems many, if not most setups on the forum, are trying to maximally reduce costs. As my application is clinical, quality is more important than cost. What is not clear is where system costs begin to escalate significantly with minimal quality increase. For example, how does one determine whether a frame grabber (what data transfer rate) is needed or not?   

The longest videos would be 10 sec at most, but as long as the subject is steady state walking, 2 second recordings should be sufficient. The data is not for research but for immediate guidance of clinical intervention so the accuracy of a Vicon motion capture system is not necessary nor desired due to extreme cost and high post-processing requirements. A quality video system is sufficient, but higher resolution and high frame rates are desirable.

The plan is to use global shutter Daheng cameras as they seem to provide the best value. I now realize that I provided the incorrect camera model number (MARS-900-120GTC). I was actually considering the MARS-561-207GTC which is also 10GigE (not GigE, but 10x faster). The maximum resolution is 2592×2160@207.77fps. The goal was to halve the vertical resolution to double the maximum possible frame rate (1296x2160@415fps), which would put me at my goal between 400-1000fps. How do I determine if two MARS-561-207GTCs require a frame grabber?  Is it based upon CPU, GPU, RAM something else? If all of the Daheng 10gigE cameras require a frame grabber, then the higher resolution USB 3.0 cameras may provide better value considering the resolution.

Regarding shutter speed, is the commonly recommended 2x fps for normal video still applicable to high speed video?  E.g., if recording at 415fps, one should have a shutter speed of 850 fps (2x) or higher to eliminate blur.

I was hoping a color camera might be more effective for simultaneously tracking multiple data points (angles & linear translations) by using differenct colored subject joint markers. If Kinovea does not distiguish color markers, then mono seems to be the better option. It may also be the difference between needing a frame grabber or not.

Thank-you for the recommendations, links, and script. I have further studying to do...

6 Reply by Reiner (edited by Reiner 2024-02-19 21:29:38)

  • Reiner
  • Member
  • Offline

Do not yet really understand what your measurements will be and for what you need such a high resolution.
I would recommend making a test using the camera of your handy at high speed and reduce the frame size afterwords to see how far you can go down to not loose the information you would like to see. I'm always wondering, of how low resolution will give a lot of information if the image is crisp enough.
Ask the representatives of Daheng regarding the technical details of the Mars camera. I think they can tell you what resolution/fps you may reach with and without a frame grabber.
I have no experience with 10GigE transfer cameras and do not know what the transfer limit will be in real life.
You may estimate the data transfer you need by using the formula: width * height * fps * C/M(factor) / 1000000, resulting in MB/s. For color use factor 3, for mono use factor 1 The limit of USB3.0 is about 350-400 MB/s.

keep in mind that reducing the HEIGHT of the image will increase the fps. Changing the width has no effect.

Blurring is NOT related to fps. Increasing the fps will increase the number of images you will capture in a given distance (in moving objects). The shutter speed alone determines, if each single image shows a blurring. The amount of blurring depends on the speed of the moving object. A shutter speed of 1000 microseconds, for example, means that for each video the camera is recording, the shutter remains 1000 microseconds open during each image is recorded. If the object moves slowly, the amount of blurring is low. If the object moves fast, also 1000 microseconds creates a significant amount of blurring. Of course, blurring is most visible, if the camera is positioned perpendicular to the moving direction.
I suggest, that your intention is to reduce the blurring in your image and a lower fps will be sufficient.

Don't know if Kinovea can recognize a colored marker. In most cases, the marker is more bright but not colored. Mono cameras have a higher light sensitivity (need less light) giving a good image contrast.

7 Reply by leroyoddie

  • leroyoddie
  • Member
  • Offline

Thank-you again for the insight! I'll reach out to Daheng to see what they recommend regarding frame grabber needs.