Perception

Section 1: Homography & Tracking

We use SIFT & ORB as feature detectors to get matching descriptors between a given image that we're matching with another image or with the live webcam.

Usage:

    detector = Detector("<query-image-path>") # images/detect.jpg as an example

detector.track_on_webcam() is used to access the webcam, it matches the descriptors and finds the homography matrix H, and applies it with a bounding box. Here's an example:

detector.detect("<target-image-path>") # e.g. - images/detect_book.jpg uses the target image to match with the query image with which the detector was initialized. After finding matching descriptors, it draws lines between them. Here's an example:

Section 2: Single Value Decomposition

We will try a Single Value Decomposition on an image with variable ranks to see how much information is being stored at each rank for image compression.

As we know, X = UΣV^T, and we want to figure out the SVD for the image i.e. X is the grayscale image

Original image ( converted to grayscale ) Size: 476KB

Rank: 5 Size: 129KB

Rank: 20 Size: 170KB

Rank: 100 Size: 187KB

Single Values

Single Values: Cumulative