MATLAB Tutorial

Computer Vision System in MATLAB

Computer Vision in MATLAB provides tools and functions to process, analyze, and interpret images and video data. MATLAB’s Computer Vision Toolbox includes pre-built functions for image processing, feature detection, object recognition, and more.

Basic Concepts of Computer Vision implemented in Matlab

Computer Vision tasks often involve image manipulation, feature extraction, and object recognition. Below are the fundamental operations:

Reading and Displaying Images

To process an image, first read it into MATLAB and then display it:


% Reading and displaying an image 

img = imread('peppers.png'); % Reads an image file 

imshow(img);                % Displays the image 

title('Original Image');

Image Processing Operations

Common image processing techniques include resizing, converting to grayscale, and applying filters:

Resizing an Image


% Resize an image 

resizedImg = imresize(img, 0.5); % Scale down the image to 50% of its size 

imshow(resizedImg); 

title('Resized Image');

Converting to Grayscale


% Convert image to grayscale 

grayImg = rgb2gray(img); 

imshow(grayImg); 

title('Grayscale Image');

Applying Filters


% Apply a Gaussian filter 

filteredImg = imgaussfilt(grayImg, 2); % Standard deviation of 2 

imshow(filteredImg); 

title('Filtered Image');

Feature Detection and Recognition

Detecting features is a critical step in Computer Vision tasks such as object tracking and recognition.

Edge Detection


% Edge detection using Canny method 

edges = edge(grayImg, 'Canny'); 

imshow(edges); 

title('Edge Detection');

Object Detection using Pre-Trained Models

MATLAB supports object detection using pre-trained deep learning models, such as YOLO and SSD.


% Load pre-trained YOLOv2 network 

net = load('yolov2ResNet50VehicleExample.mat'); 

detector = net.detector; 


% Perform object detection 

[bboxes, scores, labels] = detect(detector, img); 


% Annotate detections on the image 

detectedImg = insertObjectAnnotation(img, 'rectangle', bboxes, cellstr(labels)); 

imshow(detectedImg); 

title('Object Detection');

Working with Videos

Processing videos involves reading video frames and performing operations on each frame.


% Read a video file 

videoObj = VideoReader('traffic.mp4'); 


% Process and display each frame 

while hasFrame(videoObj) 

    frame = readFrame(videoObj); 

    imshow(frame); 

    pause(1/videoObj.FrameRate); % Pause for display 

end

Advance Example Problems on Computer Vision Systems in Matlab

Example 1: Object Tracking in Video using Optical Flow

In this example, we’ll track motion in a video using the Lucas-Kanade method for optical flow.


% Load video file 

videoObj = VideoReader('traffic.mp4'); 
 


% Read the first frame and convert to grayscale 

frame1 = rgb2gray(readFrame(videoObj)); 
 


% Initialize points to track using Harris corners 

corners = detectHarrisFeatures(frame1); 

points = corners.selectStrongest(50).Location; 
 


% Initialize optical flow tracker 

tracker = vision.PointTracker('MaxBidirectionalError', 1); 

initialize(tracker, points, frame1); 
 


% Process subsequent frames 

while hasFrame(videoObj) 

    frame2 = rgb2gray(readFrame(videoObj)); 

    [points, validity] = tracker(frame2); 

    frameWithPoints = insertMarker(frame2, points(validity, :), '+', 'Color', 'red'); 

    imshow(frameWithPoints); 

    title('Object Tracking'); 

    pause(1/videoObj.FrameRate); 

end

Explanation: This script detects Harris corners in the first frame, tracks them using optical flow, and visualizes the motion.

Example 2: Image Stitching (Panorama Creation)

Here, we’ll combine two overlapping images into a single panorama using feature matching and transformation.


% Read two overlapping images 

img1 = imread('scene1.jpg'); 

img2 = imread('scene2.jpg'); 
 


% Convert to grayscale 

gray1 = rgb2gray(img1); 

gray2 = rgb2gray(img2); 
 


% Detect and extract features 

points1 = detectSURFFeatures(gray1); 

points2 = detectSURFFeatures(gray2); 

[features1, validPoints1] = extractFeatures(gray1, points1); 

[features2, validPoints2] = extractFeatures(gray2, points2); 
 


% Match features 

indexPairs = matchFeatures(features1, features2); 

matchedPoints1 = validPoints1(indexPairs(:, 1)); 

matchedPoints2 = validPoints2(indexPairs(:, 2)); 
 


% Estimate geometric transformation 

[tform, inlierPoints1, inlierPoints2] = estimateGeometricTransform(matchedPoints1, matchedPoints2, 'projective'); 
 


% Warp the second image 

outputView = imref2d(size(img1)); 

warpedImg2 = imwarp(img2, tform, 'OutputView', outputView); 
 


% Blend the images 

panorama = max(img1, warpedImg2); 

imshow(panorama); 

title('Panorama');

Explanation: This code uses SURF features to align two overlapping images and creates a stitched panorama.

Example 3: Face Detection using Haar Cascades

Detect human faces in an image using Haar cascade classifiers.


% Load an image 

img = imread('group_photo.jpg'); 
 


% Load Haar cascade classifier 

faceDetector = vision.CascadeObjectDetector(); 
 


% Detect faces 

bboxes = step(faceDetector, img); 
 


% Annotate detected faces 

annotatedImg = insertObjectAnnotation(img, 'rectangle', bboxes, 'Face'); 

imshow(annotatedImg); 

title('Face Detection');

Explanation: The Haar cascade classifier identifies regions containing faces, and these are highlighted in the output image.

Example 4: Image Segmentation using k-Means Clustering

Segment an image into different regions based on color clustering.


% Read the input image 

img = imread('peppers.png'); 
 


% Reshape image into a 2D array of RGB values 

rows = size(img, 1); 

cols = size(img, 2); 

data = reshape(img, rows*cols, 3); 
 


% Perform k-means clustering 

k = 3; % Number of clusters 

[labels, centroids] = kmeans(double(data), k); 
 


% Reshape clustered labels back to the image dimensions 

segmentedImg = reshape(labels, rows, cols); 

imshow(label2rgb(segmentedImg)); 

title('Image Segmentation');

Explanation: This script uses k-means clustering to group pixels into k regions based on color similarity.

Example 5: Lane Detection in Driving Scenes

Detect lane lines in a driving scene using Hough transform.


% Read the driving scene image 

img = imread('road.jpg'); 
 


% Convert to grayscale and detect edges 

grayImg = rgb2gray(img); 

edges = edge(grayImg, 'Canny'); 
 


% Define a region of interest 

mask = poly2mask([50, 200, 300, 400], [400, 200, 200, 400], size(edges, 1), size(edges, 2)); 

roiEdges = edges & mask; 
 


% Perform Hough transform 

[H, T, R] = hough(roiEdges); 

lines = houghlines(roiEdges, T, R); 
 


% Draw lane lines on the image 

imshow(img); 

hold on; 

for k = 1:length(lines) 

    xy = [lines(k).point1; lines(k).point2]; 

    plot(xy(:, 1), xy(:, 2), 'LineWidth', 2, 'Color', 'yellow'); 

end
title('Lane Detection');

Explanation: The Hough transform identifies straight lines in the edge-detected image, representing lane markings.

Useful MATLAB Functions for Computer Vision

Function

Explanation

imread

Reads an image from a file.

imshow

Displays an image.

rgb2gray

Converts an RGB image to grayscale.

imresize

Resizes an image to the specified dimensions.

edge

Detects edges in an image using specified methods.

detect

Performs object detection using a pre-trained network.

Practice Questions

Test Yourself

1. Load an image and apply a Gaussian filter to it.

2. Detect edges in an image using different methods (e.g., Sobel, Canny).

3. Write a script to process each frame of a video and apply edge detection.

4. Use a pre-trained YOLO network to detect objects in an image.

5. Use optical flow to track objects across frames in a new video.

6. Segment an image of your choice into five regions using k-means clustering.

7. Detect and annotate all cars in a traffic video using Haar cascades.

8. Implement face recognition by combining face detection and a pre-trained deep learning model.