OpenQVision is a computer vision project that allows developers and/or data scientists to rapidly test their newborn neural networks on camera streaming, video or photos. As the name itself refers to, it is based on OpenCV (https://github.com/opencv/opencv) and Qt framework (https://github.com/qt) and handles Tensorflow (https://github.com/tensorflow/tensorflow) and Yolo Neural Networks (https://github.com/pjreddie/darknet).
- An up to date OpenCV
- LibCurl
- An up to date Qt Creator
- Telegram account and a bot (not mandatory)
$ git clone https://github.com/dev-gazer/OpenQVision.git$ cd OpenQVision
$ qtcreator .After building and running OpenQVision for the first time, you should see the following window:
For a quick sanity check, click on Start Camera. If it opens, it is all set.
Then click on Stop Camera.
Afterwards, browse your neural network files path (weights, config and classes) to the main window and pick your media source. You can browse to search for image files (*.png, *.jpg, .jpeg) or video files (.mp4, *.mpg, *.mpeg). If a media file is selected, the Start Camera and the Stop Camera buttons change to Read File and Stop. By pressing Read File or Start Camera, OpenQVision will identify if the user is loading a Tensorflow or Yolo model and media reproduction or camera streaming is executed coupled with the detection model.
Feel free to use OpenQVision with your own modules. This version conditions the object detection to trigger a Telegram Bot. If you want to use it, go to OpenQVision/src/curl_telegram.cpp and to OpenQVision/src/videocaptured.cpp and set your own Bot Token and Telegram User Id. Any problems with creating a Telegram Bot, please refer to official documentation at https://core.telegram.org/bots/api.
If you are using Yolo weights and config and insert a Telegram Bot Token and a Telegram User Id, the object detection will trigger the chatbot after counting five frames with an object from your model classes. The fifth frame is captured in results directory and sent with the caption "We found your !". Something like follows:
-
Alexandre Martuscelli Faria
Contact: martuscellifaria@gmail.com
-
Github C++ community for providing a way of performing get and post requests so easily in C++.
-
Tom Ha (https://github.com/tomha85) for providing such great lecture on how to read properly Yolo Darknet in C++ OpenCV.
-
Nico Nielsen (https://github.com/niconielsen32) for also providing great content on Computer Vision in C++ and Python. This encouraged me a lot on porting OpenQVision from Python to C++.
-
Audrey Siqueira (https://github.com/audrey-siqueira) for suggesting the generalization of the application, to read not just camera streaming, but also video and image files.
GNU LESSER GENERAL PUBLIC LICENSE Version 3, 29 June 2007
In any case, please refer to the LICENSE file in this folder.