Object tracking in video with OpenCV and Deep Learning

Поделиться
HTML-код
  • Опубликовано: 13 янв 2025

Комментарии • 93

  • @park2348190
    @park2348190 4 года назад +1

    I didn't know about IBM's powerai! Thanks for the info.

  • @imagenigraphics
    @imagenigraphics 4 года назад +7

    This project was explained simply but when tried, even installing IBM VIsual Insights (formerly PowerAI) is proving difficult for me.

  • @AamirKhan-px5vu
    @AamirKhan-px5vu 5 лет назад

    thanks for sharing
    such a good information

  • @pacejohn454
    @pacejohn454 5 лет назад +14

    Which model did you train with? YOLO or Faster-RCNN? I couldn't find that in the documentation or on this page. Maybe I missed it?

  • @Xpertapps
    @Xpertapps 2 года назад

    Отличная работа. Делаем что-то похожее.

  • @Fingiorstyle
    @Fingiorstyle 4 года назад

    Thank you very much for the video!!

  • @godlysunny1896
    @godlysunny1896 3 года назад

    They have made Maximo into a product for sale. How do I use it for tagging images?

  • @yosanmelese2094
    @yosanmelese2094 4 года назад +6

    I am considering counting cars worth more than 100K... and just use a highway video to estimate GDP of a country

  • @dkibe6006
    @dkibe6006 4 года назад

    is there a way to show how you can install the IBM POWER AI VISION (TRIAL VERSION ) in windows 10? am finding difficulty in downloading.thanks

    • @markstur
      @markstur  4 года назад

      The pattern was updated and Power AI Vision is now Maximo Visual Inspection. Go to the repo and see the Prerequisites section. It has this link:
      Go here to learn more about trial access (Scroll down to the Give it a try section).
      www.ibm.com/support/pages/ibm-maximo-visual-inspection

  • @aleksipesonen9831
    @aleksipesonen9831 4 года назад

    How do I make the system detect differences in "cars"? (in my case they are large paper rolls in factory). I would want the system know when there is a something odd with the object and count it

  • @lucaslin783
    @lucaslin783 4 года назад

    Hi Mark, how can I use powai vision in windows10

  • @danielmoraes9637
    @danielmoraes9637 6 лет назад +1

    great job with deep learning

  • @AndreyKostenkoDr
    @AndreyKostenkoDr 5 лет назад

    Hi Mark, in the counting_cars.ipynb, does the link to the video file need to be updated? input_video_url = "ibm.box.com/shared/static/4r4tf2robabiqu36ocmykrzvojuse5k9.mp4" Also, I cannot download the video manually using ibm.box.com/v/counting-cars-training It says: This shared file or folder link has been removed or is unavailable to you. Any suggestions?

  • @mamoudousano3354
    @mamoudousano3354 5 лет назад +1

    Fantastic tutorial, Thanks @ Mark Sturdevant. Is it possible to detect only a few cars and for a few moments, maybe cars that go very fast?

    • @markstur
      @markstur  5 лет назад +1

      If distance moved from frame to frame is good enough that would be easy to do. You'd still detect all of them, but just highlight ones that move faster. I'm guessing it would work better if you tracked the car across more frames before determining which are fast. For that you'd just have to add a little data with the tracker data to calculate the movement. That's a good idea.

    • @mamoudousano3354
      @mamoudousano3354 5 лет назад

      @@markstur ok good, thanks

  • @vidusheemowlasee7200
    @vidusheemowlasee7200 5 лет назад +2

    Hello in how much time the registration for the powerAI vision is approved?

  • @igxh5
    @igxh5 4 года назад

    Where did u get the video? Or do u know any source of more videos ?

  • @thedude333
    @thedude333 3 года назад

    hi, great video, cant access the IBM Maximo Visual Inspection though, is it still available for free ? can you please post a direct link ?
    thanks.

  • @lowqchannel
    @lowqchannel 4 года назад

    can the model differentiate car and heavy vehicle e.g. lorries, buses

    • @markstur
      @markstur  4 года назад

      you can train the model w/ images of buses, trucks, etc with different labels using Visual Insights. This quick one was trained w/ a minimal number of cars.

  • @riponsaha8320
    @riponsaha8320 4 года назад

    Great tutorial. However, I believe, you will not get better performance by training on the bigger dataset because those are auto-generated and you didn't fix any error labels induced in the auto labeling process.

  • @Cristian-ew4in
    @Cristian-ew4in 4 года назад

    Thanks for your video, bro. Do you know how to count objects in android studio?

  • @shivamverma2091
    @shivamverma2091 4 года назад

    can U share the Algorithm you used in the project?

  • @eric9964
    @eric9964 4 года назад

    So this is an api that would need to be run with wifi?

  • @sohrabi.mohammadjavad
    @sohrabi.mohammadjavad 5 лет назад +3

    i use yolov3 network for detecting , can i use this code for counting ?

    • @paulsesh
      @paulsesh 5 лет назад +3

      i have the same question

    • @ibrahimaba8966
      @ibrahimaba8966 3 года назад

      guys yolov3 is so slow...

  • @M.4.R.V.
    @M.4.R.V. 4 года назад +1

    Can I use my camera connected to my LapTop in Real-Time?

  • @miguecastillo
    @miguecastillo 4 года назад

    Hi man, do you have any idea of how far can it detect a car?

  • @TheBorlanddelphi
    @TheBorlanddelphi 4 года назад

    Hi Mark, any online class for me learn this lesson?

  • @cristiandavidparramunoz6101
    @cristiandavidparramunoz6101 4 года назад

    Nice video, do you know how to implement this in android studio?

  • @imagenigraphics
    @imagenigraphics 4 года назад

    I'm sorry but is it me or I can't find the json file?

  • @mustafauysal4187
    @mustafauysal4187 4 года назад

    PowerAI Vision changed a lot. How can I create a Dataset

  • @diegovigil7311
    @diegovigil7311 6 лет назад

    Hello Mark. I liked the video, but I have tried to do but I think that Power AI Vision has stopped working. After uploaded the video, I have got 5 Frames as you said and then Exported the Labels. However its exporting labels for 3 days. Do you know what could be happening ? Some Tip? Thank you!

    • @patrickwalsh8913
      @patrickwalsh8913 6 лет назад

      I am getting the same thing. It's been exporting for about half an hour

    • @antonioledezma1096
      @antonioledezma1096 6 лет назад

      @@patrickwalsh8913 Im also doing the same it says exporting labels with a circle on and it still the same by 10 days dunno what is going on

  • @patrickwalsh8913
    @patrickwalsh8913 6 лет назад

    I created the labels but it has been exporting for about half an hour. What did I do wrong? I drew the boxes around 5 cars from each of the two clips, but the export does not seem to work.

    • @markstur
      @markstur  6 лет назад

      Please try 5 frames and 25-30 labeled cars.

    • @patrickwalsh8913
      @patrickwalsh8913 6 лет назад

      I created 8 frames and 34 labels but it is still stuck exporting for over a week now.

    • @Late-Night-AI
      @Late-Night-AI 6 лет назад

      Got the same problem...the export of the labels doesn't work for me either

    • @markstur
      @markstur  6 лет назад

      @@Late-Night-AI I'll contact the folks maintaining SuperVessel again. Thanks for commenting in the github issue.

  • @vanhuypham7584
    @vanhuypham7584 Год назад

    too nice, thanks

  • @darshinicr3211
    @darshinicr3211 5 лет назад +1

    Sir could u explain the application of object detection and tracking

    • @markstur
      @markstur  5 лет назад +1

      Object detection allows you to identify and locate objects in an image. For example, we count the cars in an image and draw a box around each one -- instead of just classifying the picture as cars or "not cars". Tracking applies to video. We ran object detection of frames (images), but we'd like to recognize objects that we already detected. If we only cared about average number of cars at any point in time, we might not care to track them. Tracking allows us to count each car once and also time it from start to stop.

    • @darshinicr3211
      @darshinicr3211 5 лет назад

      @@markstur why only Python not c or c++

    • @markstur
      @markstur  5 лет назад

      @@darshinicr3211 You can choose another language such as c or c++. We chose Python.

  • @wijaywijayakumaran5404
    @wijaywijayakumaran5404 6 лет назад +1

    Fantastic tutorial. Is the "training_video.mp4" link missing?

    • @markstur
      @markstur  6 лет назад

      You should be able to download the file using this URL: ibm.box.com/v/counting-cars-training

    • @wijaywijayakumaran5404
      @wijaywijayakumaran5404 6 лет назад

      Thanks Mark

    • @tomtao5261
      @tomtao5261 5 лет назад

      @@markstur the url is also invaild now

    • @markstur
      @markstur  5 лет назад

      it got disabled for public access, should be fixed now

  • @anpr
    @anpr 5 лет назад

    Thanks for the video, what is the GPU used for training?

    • @akshy471
      @akshy471 5 лет назад

      NVIDIA® TITAN RTX™

    • @haroucheriadh8542
      @haroucheriadh8542 4 года назад

      @@akshy471 can we use another GPU [ex: Intel graphics Hd] ???????

    • @akshy471
      @akshy471 4 года назад

      @@haroucheriadh8542 integrated? Not much useful

    • @haroucheriadh8542
      @haroucheriadh8542 4 года назад

      @@akshy471
      Yes, unfortunately the graphic card is built-in and cannot be changed. Is there a solution ??

  • @payamb9601
    @payamb9601 4 года назад

    this is really a rookie question, but how can i use it with my own video, if the finish line is hard coded?

    • @markstur
      @markstur  4 года назад +2

      Good question. You should take the code and modify to work for you. I've seen this code leveraged in surprising ways.

  • @tahrirsiddiqui6588
    @tahrirsiddiqui6588 4 года назад

    Great tutorial! My project requires real-time object detection and motion tracking using Azure Kinect 4k camera, does OpenCV have any solutions for real-time?

    • @purplehax4957
      @purplehax4957 4 года назад

      You could try to use the YOLO v3 algorithm or an easier but not so good option would be contours.

    • @borjonx
      @borjonx 4 года назад

      I used the original Kinect (model 1414) with OpenCV inside Processing yesterday for real-time face detection. github.com/atduskgreg/opencv-processing

    • @dhruvdwivedy4192
      @dhruvdwivedy4192 4 года назад +1

      @@borjonx nIce

    • @ibrahimaba8966
      @ibrahimaba8966 3 года назад

      guys yolov3 is slow, you can just use backgroundSUbtraction and TrackerCSRT to achieve the same project in real time. I will publish a link of my code. It's fast and there is no need of deep learning for this kind of project cause the camera is static.

  • @hemanthsharmam4761
    @hemanthsharmam4761 6 лет назад +1

    I have tried kcf opencv tracker which is used in this code before.....But it used to be very slow for multiple objects....If anybody have tried this pls let know how the performance is....fps and system specs used

    • @ibrahimaba8966
      @ibrahimaba8966 3 года назад

      Hello if you sue multiple tracker, so you will need to implement mulprocess that's why your code is slow.

  • @alphaO27
    @alphaO27 5 лет назад

    How can I use this Power Ai vision of IBM ? Is it free ?

    • @markstur
      @markstur  5 лет назад

      Check here to see if a free trial is available developer.ibm.com/linuxonpower/deep-learning-powerai/vision/access-registration-form/ Otherwise you need a Check here for options developer.ibm.com/linuxonpower/deep-learning-powerai/vision/ If you don't have Power hardware then you want to look for a free trial on the cloud.

  • @viridiaromero
    @viridiaromero 6 лет назад +1

    Great project! I'm trying to run it in my anaconda prompt and it's displaying this error: File "main.py", line 80, in
    mask=cv2.morphologyEx(mask,cv2.MORPH_CLOSE,kernalCl)
    TypeError: kernel data type = 8 is not supported ... what could be the reason :/

    • @markstur
      @markstur  6 лет назад

      Thanks! I don't see morphologyEx or kernalCI used in this project. So it's hard to help, but certainly the kernalCI param is setup wrong. The "Structuring Element" section in this doc might help: docs.opencv.org/3.0-beta/doc/py_tutorials/py_imgproc/py_morphological_ops/py_morphological_ops.html

    • @rubenscj4772
      @rubenscj4772 5 лет назад

      i had same problem.

  • @yalmadiable
    @yalmadiable 5 лет назад

    Is it useful for realtime detect and track

    • @markstur
      @markstur  5 лет назад

      the shared/remote/free server isn't something you'd use for time-critical inference and that's what I've been using. If you are buying your own inference servers you should be able to get good response time (i.e. on edge devices). I know they have options, but I don't have experience with them.

    • @yalmadiable
      @yalmadiable 5 лет назад

      Well I'm currently looking to develop a thermal detection/track software it's really not easy it has to be real time

    • @markstur
      @markstur  5 лет назад

      @@yalmadiable I'm not a sales guy, but if you're looking for edge devices, this link might be the right direction www.ibm.com/us-en/marketplace/ibm-powerai-vision/details

  • @AndreyNikishaev
    @AndreyNikishaev 5 лет назад

    You can master Object Detection in this specialized practical online course: learnml.today

  • @ALLINONETrending
    @ALLINONETrending 4 года назад

    Hello bro...this video is awesome...i wanted to recognise a spect on a person eye and vehicle name ..is it possible. !?? Please suggest me how to do that ?

  • @stizandelasage
    @stizandelasage 4 года назад +4

    i'm turning this into a weapon

  • @khoalt8216
    @khoalt8216 5 лет назад

    Can you share for me origin video ?! Please :))

    • @markstur
      @markstur  5 лет назад

      training and testing videos are here github.com/IBM/powerai-counting-cars/tree/master/data

  • @user-yw8sr3uj1w
    @user-yw8sr3uj1w 5 лет назад

    mac user