Vision Basics

How Cameras Work in FTC

Your robot's camera captures frames at a specific resolution (typically 640x480 or 1280x720) and frame rate. Each frame is a 2D array of pixels, where each pixel contains color information.

Color Spaces

Different color spaces represent colors in different ways. Understanding them is crucial for effective color detection.

RGB (Red, Green, Blue)

The most common color space. Each pixel has three values (0-255) representing red, green, and blue intensity.

// RGB values
int red = 255;
int green = 0;
int blue = 0;
// Result: Pure red color

Pros: Intuitive, widely used
Cons: Not ideal for color detection due to lighting sensitivity

HSV (Hue, Saturation, Value)

HSV separates color information (hue) from brightness (value) and intensity (saturation).

• Hue (0-179 in OpenCV) - The actual color (red, green, blue, etc.)
• Saturation (0-255) - How "pure" the color is
• Value (0-255) - How bright the color is

// HSV ranges for detecting red
Scalar lowerRed = new Scalar(0, 100, 100);
Scalar upperRed = new Scalar(10, 255, 255);

Pros: Much better for color detection, lighting-independent
Cons: Slightly less intuitive

YCrCb (Luma, Chroma Red, Chroma Blue)

YCrCb separates brightness (Y) from color information (Cr and Cb channels).

• Y (0-255) - Brightness/luminance
• Cr (0-255) - Red chrominance
• Cb (0-255) - Blue chrominance

// YCrCb is great for detecting specific alliance colors
Scalar lowerBlue = new Scalar(0, 0, 150);
Scalar upperBlue = new Scalar(255, 120, 255);

Pros: Excellent for FTC alliance color detection
Cons: Less intuitive than HSV

Common Vision Techniques

Thresholding

Convert an image to binary (black and white) based on color ranges. Pixels within range become white, others become black.

// Create a mask showing only red pixels
Core.inRange(inputMat, lowerBound, upperBound, mask);

Contour Detection

Find the boundaries of white regions in a binary image. Useful for identifying game elements.

List<MatOfPoint> contours = new ArrayList<>();
Imgproc.findContours(mask, contours, hierarchy, 
    Imgproc.RETR_EXTERNAL, Imgproc.CHAIN_APPROX_SIMPLE);

Bounding Boxes

Draw rectangles around detected objects to determine their position and size.

Rect boundingRect = Imgproc.boundingRect(contour);
// Now you know: x, y, width, height

Region of Interest (ROI)

Analyze only specific parts of the image to improve performance and reduce false positives.

Mat leftRegion = inputMat.submat(
    new Rect(0, 240, 213, 240) // x, y, width, height
);

Performance Considerations

Resolution vs Speed

• Lower resolution (320x240) - Faster processing, less detail
• Higher resolution (1280x720) - More detail, slower processing

Choose based on your needs. Most teams use 640x480 or 640x360 as a good balance.

Frame Rate

• Higher FPS - More responsive, more CPU usage
• Lower FPS - Less responsive, less CPU usage

Typical FTC cameras run at 30 FPS, but you may process every 2nd or 3rd frame to save resources.

Processing Location

• OnBot Java/Kotlin - Runs on Control Hub, simpler deployment
• Android Studio - Runs on Robot Controller phone, more power but deprecated

Testing Your Vision Code

EOCV-Sim

EOCV-Sim is an offline simulator where you can:

• Test pipelines without a robot
• Use images or webcam input
• Tune parameters quickly
• See processing results in real-time

See EOCV-Sim Setup for details.

FTC Dashboard

FTC Dashboard allows you to:

• Stream camera feed over WiFi
• Adjust parameters in real-time
• View telemetry alongside video
• Save tuned values back to code

Next Steps

Now that you understand the basics:

• Learn EasyOpenCV setup for custom pipelines
• Or jump to VisionPortal for quick implementation
• Explore EOCV-Sim for offline testing