MATLAB Code For Video Pattern Matching

Introduction:

In this code you use normalized cross correlation to track a target pattern in a video. The pattern matching algorithm involves the following steps:

Initialize Parameters and Create a Template:

Initialize required variables such as the threshold value for the cross correlation and the decomposition level for Gaussian Pyramid decomposition.

Prepare a video file reader.

Specify the target image and number of similar targets to be tracked. By default, the example uses a predefined target and finds up to 2 similar patterns. You can set the variable useDefaultTarget to false to specify a new target and the number of similar targets to match.

Create a System object to calculate the local maximum value for the normalized cross correlation.

Create a System object to display the tracking of the pattern.

Initialize figure window for plotting the normalized cross correlation value

Search for a Template in Video

Create a processing loop to perform pattern matching on the input video. This loop uses the System objects you instantiated above. The loop is stopped when you reach the end of the input file, which is detected by the VideoReader object.

Summary:

This code shows use of Computer Vision Toolbox™ to find a user defined pattern in a video and track it. The algorithm is based on normalized frequency domain cross correlation between the target and the image under test. The video player window displays the input video with the identified target locations. Also a figure displays the normalized correlation between the target and the image which is used as a metric to match the target. As can be seen whenever the correlation value exceeds the threshold (indicated by the blue line), the target is identified in the input video and the location is marked by the green bounding box.

See the Video:

Click here.