I require your help regarding a problem I’m encountering.
Description:
Hardware concerned:
Software concerned:
Troubleshooting steps already taken:
Additional information:
Thank you so much in advance for your help!
Hello all.
concerns: football.
Camera Type:
Sunba 1080P Outdoor Auto Tracking 25x Optical Zoom Ip Poe + Ptz camera, rtmp and built-in microphone, 300m IR night vision (Illumination)
for your information:
I have this camera hanging on top of the grandstand (height 15 meters) Netherlands
what am I looking for:
I am looking for a software or a kind of robot under the camera with a built-in laser with which the camera / tracking can follow the football (white ball).
The camera in question would probably be sufficient to detect the ball, assuming you connect it to something that can process the video stream and analyze the images fast enough ( insert machine learning! ). To detect the position related to the field/players/other objects it would probably be best to have two cameras offset from each other slightly to help with accurate depth detection (like your eyes!).
What would that be for?
You’ll need some hardware that is fast at processing video data and algorithms to detect objects in them. And, of course, the software to do so. You may want to look into what OpenCV can do for you (or some other computer vision library/software).
If you need to track the ball in real time you’ll probably need some beefy hardware (something like the nvidia jetson modules might be a good fit for this). On the other hand, if you can simply record all of the footage and analyze this later then it can go as slow as you want!
Sunba 1080P Outdoor Auto Tracking 25x Optical Zoom Ip Poe + Ptz camera, rtmp and built-in microphone, 300m IR night vision (Illumination)
The camera in question would probably be sufficient to detect the ball, assuming you connect it to something that can process the video stream and analyze the images fast enough ( insert machine learning! ). To detect the position related to the field/players/other objects it would probably be best to have two cameras offset from each other slightly to help with accurate depth detection (like your eyes!).
hey Scharette am glad you respond! The Sunba camera pleases with regard to image quality very well. only the speed of tracking the ball cannot cope with this tracking. we have a YouTube channel where we stream the live images Buying a second Sunba camera is no problem what kind of machine learning should I apply (do you happen to have a link?)
Edmund:
with a built-in laser
What would that be for?
no idea why a laser came up. perhaps because a laser can reach a great distance. related to following a football. but if it is not needed by the software then not. I don’t understand the software that can do it.
Edmund:
software or a kind of robot
You’ll need some hardware that is fast at processing video data and algorithms to detect objects in them. And, of course, the software to do so. You may want to look into what OpenCV can do for you (or some other computer vision library/software).
If you need to track the ball in real time you’ll probably need some beefy hardware (something like the nvidia jetson modules might be a good fit for this). On the other hand, if you can simply record all of the footage and analyze this later then it can go as slow as you want!
Imagine that I control two Sunba cameras, can you indicate which hardware and software I need so that I can follow the (white) football at high speed?
You are talking about nvidia jetson modules I will have to ask others because I know too little about them! I so hope you can link me the right hardware and software so that I can only install it on the server. what does this hardware and software cost. again thank you very much greetings Edmund
Lasers often come up when mentioning measuring distances (often used for very long range) and detecting/tracking objects (yes, but not as much as people think!
Well, you’ll need some kind of platform that is both powerful enough to do the video processing of those two cameras in real time and also analyze the content to find the ball and track it.
Definitely have a look at the material available on nvidia’s website. There’s plenty of info there. You can also start trying things out on a regular desktop or laptop with a decently beefy GPU.
Well, pretty much any modern PC hardware can do some object detection/recognition/tracking. The quality of the hardware and software will of course change how well and how fast this happens.
This will vary greatly on what you decide to use, where you buy it from and when. As said above, though, you can start exploring by using videos you already have and most likely PC hardware you already own.
I recommend trying things out and seeing if you can make it work (doesn’t need to be in real-time yet). Once you have a working solution you should be able to identify by then where are the bottlenecks and upgrade hardware as needed instead of trying to guess beforehand. Depending on how your computer vision is achieved it may be as simple as getting a better GPU.
OpenCV is definitely a place where you want to check out tutorials and documentation. There is a lot of info there about computer vision. And you’ll want to look at what has been already done, too, such as this article. TensorFlow is definitely something else you’ll want to have a look at.
Well, you have quite a bit of reading ahead of you…! Good learning!
I checked the camera you linked on Amazon and although it looks like a great solution for intruder tracking, doesn’t seem to be the best for custom object tracking as it only offers three tracking modes (line cross, intrusion and preset) all of which track any moving object. It doesn’t seem like it allows only tracking a specific object (like the football white ball). So I think it is better to opt for another option so as not to “waste” the capabilities of that camera for the task you mention.
Some options I can think of are to use a smart camera that allows you to track a specific object, you can find some options here.
Or if you feel comfortable programming the options that scharette mentioned are great, which in a few words are to use any camera that you consider fits your task (resolution/FPS/FOV) and use a SCB (like a Raspberry Pi or even better a Jetson module) to analyze the frames. There are many algorithms to detect objects either based on color or even more robust Deep Learning solutions that you can use to track any type of object. If you are interested in that, as scharette suggested, check OpenCV documentation and tutorials
Good day friends, I want to share, maybe someone was attracted by such a decision.
Not long ago I developed a real and software prototype (the Real time FSM included in the framework) of color recognition by counting pixels in a given area. During the experiments, it turned out that the method allows identifying some objects with the correct geometry.
You will need a high resolution USB or RS232 camera (maybe a microscope, etc.) to work.
The module recognizes colors, gradations and geometric (2D) colors in the designated area. The coordinates of the region can be read. Also for parametric software experiments for accurate calibration. In addition, you have the ability to manage events (open logic outputs on 16 channels).
You can also set the scan time and frequency.
The calibration procedure allows you to adapt the system to ambient lighting, camera and video card parameters, as well as to the properties of the material surface.
In the video, it is incredibly noticeable how the camera was catching yellow pixels at the border of the red marker, which is the result of the glare of the LED lamp.
). To detect the position related to the field/players/other objects it would probably be best to have two cameras offset from each other slightly to help with accurate depth detection (like your eyes!).
Edmund:
). To detect the position related to the field/players/other objects it would probably be best to have two cameras offset from each other slightly to help with accurate depth detection (like your eyes!).