Interaction Design WikiComputer Vision

Einfache Algorithmen

Auf dieser Seite werden ein paar einfach Computer Vision Algorithmen gezeigt und erklärt.

Background Subtraction

Wollen wir beispielsweise Bewegung auf einem Bild erkennen, so ist es hilfreich eine Art “Green-Screen” zu haben, welcher dann nur die Bereiche zeichnet, welche sich verändern. Dies erreichen wir mit einem sog. Background-Removal oder Background-Substraction. Dazu vergleichen wir ein statisches Referenzbild mit einem Live-Bewegtbild. Wenn die Differenz zwischen beiden Bildern (und damit die Differenz der Farbwerte an der Selben Stelle) grösser ist als ein bestimmter Schwellenwert, können wir davon ausgehen, dass an diesem Punkt Bewegung stattfindet. Leider muss man als Abstrich sagen, dass die automatische Helligkeitskorrektur der iSight Kamera sich schlecht für diese Methode auswirkt.

Beispiel Expand source
import processing.video.*;
Capture video;
 
PImage backgroundImage;
float threshold = 20;
 
void setup() {
  size(640, 480);
  
  // start video captire
  video = new Capture(this, width, height, 30);
  video.start();
  
  // prepare image to save background
  backgroundImage = createImage(video.width, video.height, RGB);
}
 
void draw() {
  // read camera image if available
  if (video.available()) {
    video.read();
  }
  
  // active pixel manipulation of canvas
  loadPixels();
  
  // get pixel data from video and background image
  video.loadPixels();
  backgroundImage.loadPixels();
   
  // loop through video pixel by pixel
  for (int x=0; x < video.width; x++) {
    for (int y=0; y < video.height; y++) {
      // get pixel array location
      int loc = x + y * video.width;
      
      // get foreground color (video)
      color fgColor = video.pixels[loc];
      
      // get background color (image)
      color bgColor = backgroundImage.pixels[loc];
       
      // get individual colors
      float r1 = red(fgColor);
      float g1 = green(fgColor);
      float b1 = blue(fgColor);
      float r2 = red(bgColor);
      float g2 = green(bgColor);
      float b2 = blue(bgColor);
      
      // calculate spacial distance between the two colors
      float dist = dist(r1, g1, b1, r2, g2, b2);
       
      // check if distance is above threshold
      if (dist > threshold) {
        // write foreground pixel
        pixels[loc] = fgColor;
      } else {
        // set pixel to black
        pixels[loc] = color(0);
      }
    }
  }
 
  // write pixel back to canvas
  updatePixels();
}
 
void mousePressed() { 
  // copy current video frame into background image
  backgroundImage.copy(video, 0, 0, video.width, video.height, 0, 0, video.width, video.height);
  backgroundImage.updatePixels();
}

Hellster Punkt

Für die direkte Steuerung eines Interfaces kann es hilfreich sein, zu wissen, wo sich der hellste Punkt in einem Bild befindet. Dazu wird ein PVector erstellt und eine Variable, welche den jeweils hellsten Wert für das aktuelle Frame beinhaltet. Durch das Vergleichen der Helligkeitswerte im ganzen Frame kann sehr schnell der Hellste Punkt bestimmt werden.

Beispiel Expand source
import processing.video.*;
Capture video;

void setup() {
  size(640, 480);

  // start video capture
  video = new Capture(this, width, height, 30);
  video.start();
}

void draw() {
  // read new video frame if available
  if (video.available()) {
    video.read();
  }

  // initially set brightness to zero
  float brightness = 0;

  // initially set point to center
  PVector point = new PVector(width/2, height/2);

  // go through video pisel by pixel
  for (int x=0; x < width; x++) {
    for (int y=0; y < height; y++) {
      // get pixel location
      int loc = x + y * width;

      // get color of pixel
      color c = video.pixels[loc];

      // check if brightness is higher than current value
      if (brightness(c) > brightness) {
        // set new brightness
        brightness = brightness(c);

        // save location of brighter point
        point.x = x;
        point.y = y;
      }
    }
  }

  // draw video
  image(video, 0, 0);

  // draw circle
  ellipse(point.x, point.y, 20, 20);
}

Farbtracking

Das Farbtracking ist eine sehr einfach Method um ein farbiges Objekt in einem Bild zu finden. Dazu wird einfach der Punkt im Bild gesucht der der festgelegtern Farbe am ähnlichsten ist.

Beispiel Expand source
import processing.video.*;
Capture video;

color trackColor;

void setup() {
  size(640, 480);

  // start video capture
  video = new Capture(this, width, height, 15);
  video.start();

  // initialize track color to red
  trackColor = color(255, 0, 0);
}

void draw() {
  // read video frame if available
  if (video.available()) {
    video.read();
  }

  // load pixels
  video.loadPixels();

  // draw video
  image(video, 0, 0);

  // initialize record to number greater than the diagonal of the screen
  float record = width+height;

  // initialize variable to store closest point
  PVector closestPoint = new PVector();
  
  // get track color as vector
  PVector trackColorVec = new PVector(red(trackColor), green(trackColor), blue(trackColor));

  // go through image pixel by pixel
  for (int x=0; x < video.width; x++) {
    for (int y=0; y < video.height; y++) {
      // get pixel location
      int loc = x + y * video.width;
      
      // get pixel color
      color currentColor = video.pixels[loc];

      // get current color as vector
      PVector currColorVec = new PVector(red(currentColor), green(currentColor), blue(currentColor)); 
      
      // calculate distance between current color and track color
      float dist = currColorVec.dist(trackColorVec);

      // save point if closer than previous
      if (dist < record) {
        record = dist;
        closestPoint.x = x;
        closestPoint.y = y;
      }
    }
  }

  // draw point if we found a one that is less than 10 apart
  if (record < 10) {
    fill(trackColor);
    strokeWeight(4.0);
    stroke(0);
    ellipse(closestPoint.x, closestPoint.y, 50, 50);
  }
}

void mousePressed() {
  // save color of current pixel under the mouse
  int loc = mouseX + mouseY * video.width;
  trackColor = video.pixels[loc];
}

Blob Detection

Die Blob Detection is schon ein komplexere Art von Algorithm, wo ein gesamtes Objekt (Blop) zu erkennen versucht wird.

Beispiel Expand source
import processing.video.*;

Capture video;

// the color to track
color trackColor;

// a dimensional array to store marked pixels
boolean marks[][];

// the total marked pixels
int total = 0;

// the most top left pixel
PVector topLeft;

// the most bottom right pixel
PVector bottomRight;

void setup() {
  size(640, 480);

  // start video capture
  video = new Capture(this, width, height, 15);
  video.start();

  // set initial track color to red
  trackColor = color(255, 0, 0);

  // initialize marks array
  marks = new boolean[width][height];
}

void draw() {
  // read video frame if available
  if (video.available()) {
    video.read();
  }

  // draw video image
  image(video, 0, 0);

  // find track color with treshold
  findBlob(20);

  // load canvas pixels
  loadPixels();

  // draw blob
  for (int x = 0; x < width; x ++ ) {
    for (int y = 0; y < height; y ++ ) {
      // get pixel location
      int loc = x + y*width;

      // make pixel red if marked
      if (marks[x][y]) {
        pixels[loc] = color(255, 0, 0);
      }
    }
  }

  // set canvas pixels
  updatePixels();

  // draw bounding box
  stroke(255, 0, 0);
  noFill();
  rect(topLeft.x, topLeft.y, bottomRight.x-topLeft.x, bottomRight.y-topLeft.y);
}

void mousePressed() {
  // save current pixel under mouse as track color
  int loc = mouseX + mouseY*video.width;
  trackColor = video.pixels[loc];
}

void findBlob(int threshold) {
  // reset total
  total = 0;

  // prepare point trackers
  int lowestX = width;
  int lowestY = height;
  int highestX = 0;
  int highestY = 0;

  // prepare track color vector
  PVector trackColorVec = new PVector(red(trackColor), green(trackColor), blue(trackColor));

  // go through image pixel by pixel
  for (int x = 0; x < width; x ++ ) {
    for (int y = 0; y < height; y ++ ) {
      // get pixel location
      int loc = x + y*width;

      // get color of pixel
      color currentColor = video.pixels[loc];

      // get vector of pixel color
      PVector currColorVec = new PVector(red(currentColor), green(currentColor), blue(currentColor));

      // get distance to track color
      float dist = currColorVec.dist(trackColorVec);
      
      // reset mark
      marks[x][y] = false;

      // check if distance is below threshold
      if (dist < threshold) {
        // mark pixel 
        marks[x][y] = true;
        total++;

        // update point trackers 
        if (x < lowestX) lowestX = x;
        if (x > highestX) highestX = x;
        if (y < lowestY) lowestY = y;
        if (y > highestY) highestY = y;
      }
    }
  }

  // save locations
  topLeft = new PVector(lowestX, lowestY);
  bottomRight = new PVector(highestX, highestY);
}

Weitere Informationen