0% found this document useful (0 votes)
9 views3 pages

CV Quiz1 Solutions

The document contains solutions to a quiz on Computer Vision, addressing various questions related to perspective projection, semantic understanding limitations, stereo imaging setups, epipolar constraints, camera calibration, and edge detection. Each question is assigned to a teaching assistant who provides detailed explanations and mathematical equations relevant to the topic. The solutions highlight key concepts such as depth calculation, the fundamental matrix, and the characteristics of image patches in relation to edges and corners.

Uploaded by

Pratyush Harsh
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
9 views3 pages

CV Quiz1 Solutions

The document contains solutions to a quiz on Computer Vision, addressing various questions related to perspective projection, semantic understanding limitations, stereo imaging setups, epipolar constraints, camera calibration, and edge detection. Each question is assigned to a teaching assistant who provides detailed explanations and mathematical equations relevant to the topic. The solutions highlight key concepts such as depth calculation, the fundamental matrix, and the characteristics of image patches in relation to edges and corners.

Uploaded by

Pratyush Harsh
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 3

Computer Vision

Spring 2024

Quiz-1 Solutions

Question 1
Assigned TA: Sreenya Chitluri
Solution:
Parallel lines in the real world appear to converge to a point on the image plane due
to perspective projection. Perspective projection is a phenomenon where objects appear
smaller as they move further away from the observer. This effect causes parallel lines,
which are actually equidistant from each other in physical space, to appear to converge
towards a vanishing point in the distance when projected onto a two-dimensional surface,
such as a photograph or a canvas.
Example - Parallel railway tracks intersect in an image.

Other logical or mathematical reasonings have also been accepted if they are
relevant and complete.

Question 2
Assigned TA: Sreenya Chitluri
Solution:
1. Limited Semantic Understanding
2. Loss of Spatial Information
3. Vocabulary Size and Generalization
Stating the reason and explaining it briefly is expected. Other reasons which
have not been stated above and are correct are also accepted.

Question 3
Assigned TA: Mohd Hozaifa Khan
Solution:
We can exploit a stereo imaging setup and use the concept of disparity to calculate depth.
However, with a textureless wall, correspondence finding methods using wall features
are ineffective. Instead, we can use our main camera as a source of unique features.
We position two additional cameras facing the main camera on the wall, creating a parallel
stereo camera setup, as shown in 1. Using the equation
f ·b
z=
d
CV: Quiz-1 Solutions

Figure 1: Q3-Stereo Setup

, we can compute depth, where f is the focal length, b is the baseline (the distance between
the optical centres of the two cameras), and d is disparity.
Disparity computation involves solving a correspondence matching problem to find corre-
sponding points in different images. For this, we can use a template or feature-matching
method, with the search space limited to a line due to our parallel setup. We can calculate
disparity using the equation d = (xl − xr ), where xl and xr are the x-coordinates of the
corresponding points in the left and right images, respectively.
By using our main camera as the source of distinct features, we can accurately compute
depth in our stereo imaging setup, even with a textureless wall.

In case of alternate solutions, the following criteria will be used:

• Clarity and completeness of solution

• Violation of constraints given in the question

• Correctness of solution given your assumptions

Question 4
Assigned TA: Sanyam
Solution:
Definition of Epipolar Constraint with Fundamental Matrix equation (0.5): "We have
epipolar plane P created using baseline B and ray R1. e1 and e2 are epipoles, and L2
is the epipolar line. Based on the epipolar geometry, search space for pixel in image i2
corresponding to pixel x1 is constrained to a single 2D line which is the epipolar line l2.
This is called the epipolar constraint."
Explanation for Fundamental Matrix Equation: xT1 F x2 = 0
Examples (0.5): Reduced Search Space, Stereo Disparity, Parallel Images (Any one)

Question 5
Assigned TA: Sanyam
Solution:
Part 1 (0.5): Rectangular Pixels/Individual Sensors

2
CV: Quiz-1 Solutions

Part 2 (0.5): Calibration Matrix with explicit explanation of fx and fy (and how it is
related to magnification) The focal length fx (for example) is actually the product of the
physical focal length of the lens and the size sx of the individual imager elements (this
should make sense because sx has units of pixels per millimeter† while F has units of
millimeters, which means that fx is in the required units of pixels). Of course, similar
statements hold for fy and sy .
fx = F ∗ sx and fy = F ∗ sy

Question 6
Assigned TA: Brunda
Solution:
0.5 - For writing the assumptions where we can write 3*4 projection matrix as 3*3 0.5 -
For writing example in the real world.
Approach 1: Assuming the object is planar (say xy plane and Z=0), then we can write
the 3*4 projection matrix as 3*3 by eliminating the third column of the matrix Example:
Capturing a planar wall (any relevant example)
Approach2: If the translation component is zero in the projection or when the camera
coordinates and world coordinates are same or the camera is assumed to be at the world
origin.(Any Relevant Example)

Question 7
Assigned TA: Shreya
Solution:
For a patch P in an image, let λ1 and λ2 be the largest and smallest eigen values of H,
and x1 and x2 be the corresponding eigen vectors.
Here x1 and x2 are the directions of the largest and smallest intensity gradients respec-
tively. And λ1 and λ2 are the corresponding magnitudes of gradients.
If the patch P contains an edge, the gradient of intensity along the edge will be zero or
close to zero, and the gradient perpendicular to the edge will be high. Hence, λ2 will be
small (close to zero), and λ1 will be large.
A corner consists of two non-parallel edges. Hence, a significant directional gradient will
exist in all directions. i.e. both λ1 and λ2 will be large and comparable.
Hence, to check if a P contains an edge, or a corner, we can check the following:

• if λ1 ∼ λ2 and λ1 , λ2 ≫ 0 then P contains a corner.

• else if λ1 ≫ λ2 and λ2 ∼ 0 then P contains an edge.

• else if λ1 , λ2 ∼ 0 then P contains a flat surface.

You might also like