0% found this document useful (0 votes)
7 views14 pages

Overview of ICP Algorithm

The document presents an overview of the Iterative Closest Point (ICP) algorithm, highlighting its ability to align 3D point clouds without manual input and its reliability in handling noisy data and partial overlaps. It outlines the steps involved in the ICP process, including initialization, finding correspondences using k-d trees, and checking for convergence. Additionally, it discusses competitive analysis and future roadmap for the algorithm's development.

Uploaded by

Sumiya AFROSE
Copyright
© All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
7 views14 pages

Overview of ICP Algorithm

The document presents an overview of the Iterative Closest Point (ICP) algorithm, highlighting its ability to align 3D point clouds without manual input and its reliability in handling noisy data and partial overlaps. It outlines the steps involved in the ICP process, including initialization, finding correspondences using k-d trees, and checking for convergence. Additionally, it discusses competitive analysis and future roadmap for the algorithm's development.

Uploaded by

Sumiya AFROSE
Copyright
© All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd

P R E S E N TAT I O N O N

Iterative Closest Point (ICP)

Present to Presented By
Dr Mehedi Hassan
Algorithm Pratik Barua [Roll: 2499 ]
Assistant Professor, Sumiya Afrose [Roll: 2491]
Dept of RME, DU Dept. of RME, DU

Department of Robotics and Mechatronics Engineering


Problem before ICP
01 02 03

Problem 1 Problem 2 Problem 3


Early methods required Feature-based Traditional methods
manual or known point algorithms failed on struggled with noisy
matches, which were dense or textureless data, outliers, or when
unavailable in real- point clouds with no point clouds only
world 3D scans clear keypoints partially overlapped
What ICP Feature 1
Identifies closest points

DOES? between source and


target clouds without
manual input

Feature 2 Feature 3 Feature 4


Computes the best Continuously improves Works reliably on dense,
rotation and translation to accuracy through multiple unstructured point clouds
align point clouds matching–transform —even with partial
precisely cycles until convergence overlaps or outliers
Iteratively Match and
01 Align
ICP repeatedly finds the closest point
correspondences and computes the best-
fit rigid transformation

Apply and Update


02 The transformation is applied to the
source points, and alignment error is
recalculated.

Check for Convergence


03
Pipeline of The process continues until the error falls
below a threshold, indicating proper

ICP alignment
Step 1: Initialization
ICP doesn’t start from scratch; it begins with an initial estimate of the
transformation R_0, t_0 that roughly aligns the source cloud P to the
target cloud Q

Here P_0 represents the source cloud after applying the initial transformation. The
quality of initial guess is crucial; a poor initialization can lead ICP to converge to an
incorrect algorithm (a local minima) or fail to converge at all.
Step 2: Finding Correspondences
For each point p_i,k in the current iteration’s transformed source cloud P_k find the
closest point q_i,k in the target point cloud Q. “Closest” is typically defined by the
Euclidean distance. This steps establishes a set of relevant corresponding pairs p_i,k
, q_i,k)

However finding matching pairs for every point in the cloud is computationally expensive. To
overcome this bottleneck, the most common approach is to build a k-d tree on static target
cloud Q. A k-d tree partitions the 3D space recursively, allowing for efficient nearest neighbor
search. This results in significant speeding the process as it reduces the search space by
finding nearest neighbors of query point from P_k.
Step 2: Finding Correspondences
For each point p_i,k in the current iteration’s transformed source cloud P_k find the
closest point q_i,k in the target point cloud Q. “Closest” is typically defined by the
Euclidean distance. This steps establishes a set of relevant corresponding pairs p_i,k
, q_i,k)

However finding matching pairs for every point in the cloud is computationally expensive. To
overcome this bottleneck, the most common approach is to build a k-d tree on static target
cloud Q. A k-d tree partitions the 3D space recursively, allowing for efficient nearest neighbor
search. This results in significant speeding the process as it reduces the search space by
finding nearest neighbors of query point from P_k.
Conclusion

Case 1 Case 2 Solution


If we knew the correct
If we knew the correct Since we know neither,
point correspondences,
transformation, finding ICP alternates between
computing the
correspondences would matching and
transformation would
also be easy transforming until
be easy.
convergence
Competitive
Analysis
Your main Competitors and their products, how are they different
and what value do you provide to your customers?

Give a striking Give a striking Give a striking Give a striking


overview of the overview of the overview of the overview of the
problem and explain problem and explain problem and explain problem and explain
it briefly. it briefly. it briefly. it briefly.
Insight 1
01 Describe how you envision to solve the
problems you previously shared.

Insight 2
02 Describe how you envision to solve the
problems you previously shared.

Insight 3
03 Describe how you envision to solve the
problems you previously shared.
Elegant
01 Describe how you envision to solve the
problems you previously shared.

Stylish
02 Describe how you envision to solve the
problems you previously shared.

Attractive
03 Describe how you envision to solve the
problems you previously shared.
Future Roadmap
What are your next steps and goals? How much support do you
need from investors and what will it get you?

Prototyp Launch V2.0 V2.2


Q1 2022 Q2 2022 Q3 2022 Q4 2022
e
Thank You FOR YOUR TIME

Fauget Inc. Alfredo Torres hello@[Link]

You might also like