J2S1-Kobuki/docs/code/OpenCV.md

# OpenCV
## Requirements
We want that the camera we want it to detect what is happening on the video feed and identify it so it can identify dangers.    

## Issues
* OpenCL not grabbing gpu
* Solution: https://github.com/Smorodov/Multitarget-tracker/issues/93

## Installation
### Dependencies
* glew (for openGL)
* opencv C++ lib

How to install OpenCV
```bash
 sudo apt-get install libopencv-dev
```

## Code explanation

### Opening the camera with OpenCV
```cpp
    VideoCapture cap(0);     //Open the default camera (0), points to /dev/video0. You could also change the number to the preferred camera
    if (!cap.isOpened()) { //if camera is not opened throw a error message
        cerr << "Error: Could not open camera" << endl;
        return;
    }
```

## Taking a picture and storing it in a variable
```cpp
    Mat frame; //create a new Matrix variable called frame
    while (true) {
        cap >> frame; // Capture a new image frame. 
        if (frame.empty()) { //if the variable frame is not filled return a error
            cerr << "Error: Could not capture image" << endl;
            continue;
        }
```

## Encoding the image for sending it over MQTT
```cpp
        vector<uchar> buf; //create a dyanmic buffer for the image 
        imencode(".jpg", frame, buf); //encode the image to the buffer
        auto* enc_msg = reinterpret_cast<unsigned char*>(buf.data());
```

```cpp

void CapnSend() {




        // Convert the image to a byte array


        // Publish the image data
        client.publishMessage("kobuki/cam", string(enc_msg, enc_msg + buf.size()));
        cout << "Sent image" << endl;

        std::this_thread::sleep_for(std::chrono::milliseconds(300)); // Send image every 1000ms
    }
}
```


## Sources
* https://github.com/UnaNancyOwen/OpenCVDNNSample/tree/master