HyperAIHyperAI

Command Palette

Search for a command to run...

a month ago

Go-ICP: A Globally Optimal Solution to 3D ICP Point-Set Registration

Yang Jiaolong Li Hongdong Campbell Dylan Jia Yunde

Go-ICP: A Globally Optimal Solution to 3D ICP Point-Set Registration

Abstract

The Iterative Closest Point (ICP) algorithm is one of the most widely usedmethods for point-set registration. However, being based on local iterativeoptimization, ICP is known to be susceptible to local minima. Its performancecritically relies on the quality of the initialization and only localoptimality is guaranteed. This paper presents the first globally optimalalgorithm, named Go-ICP, for Euclidean (rigid) registration of two 3Dpoint-sets under the L2 error metric defined in ICP. The Go-ICP method is basedon a branch-and-bound (BnB) scheme that searches the entire 3D motion spaceSE(3). By exploiting the special structure of SE(3) geometry, we derive novelupper and lower bounds for the registration error function. Local ICP isintegrated into the BnB scheme, which speeds up the new method whileguaranteeing global optimality. We also discuss extensions, addressing theissue of outlier robustness. The evaluation demonstrates that the proposedmethod is able to produce reliable registration results regardless of theinitialization. Go-ICP can be applied in scenarios where an optimal solution isdesirable or where a good initialization is not always available.

Benchmarks

BenchmarkMethodologyMetrics
point-cloud-registration-on-3dmatch-at-least-1Go-ICP
Recall (0.3m, 15 degrees): 22.9
point-cloud-registration-on-eth-trained-onGO-ICP
Recall (30cm, 5 degrees): 1.54
point-cloud-registration-on-fp-o-eGO-ICP
RRE (degrees): -
RTE (cm): -
Recall (3cm, 10 degrees): 0.00
point-cloud-registration-on-fp-o-hGO-ICP
RRE (degrees): -
RTE (cm): -
Recall (3cm, 10 degrees): 0.00
point-cloud-registration-on-fp-o-mGO-ICP
RRE (degrees): -
RTE (cm): -
Recall (3cm, 10 degrees): 0.00
point-cloud-registration-on-fp-r-eGO-ICP
RRE (degrees): -
RTE (cm): -
Recall (3cm, 10 degrees): 0.00
point-cloud-registration-on-fp-r-hGO-ICP
RRE (degrees): -
RTE (cm): -
Recall (3cm, 10 degrees): 0.00
point-cloud-registration-on-fp-r-mGO-ICP
RRE (degrees): 1.87
RTE (cm): 2.71
Recall (3cm, 10 degrees): 0.06
point-cloud-registration-on-fp-t-eGO-ICP
RRE (degrees): -
RTE (cm): -
Recall (3cm, 10 degrees): 0.00
point-cloud-registration-on-fp-t-hGO-ICP
RRE (degrees): -
RTE (cm): -
Recall (3cm, 10 degrees): 0.00
point-cloud-registration-on-fp-t-mGO-ICP
RRE (degrees): -
RTE (cm): -
Recall (3cm, 10 degrees): 0.00

Build AI with AI

From idea to launch — accelerate your AI development with free AI co-coding, out-of-the-box environment and best price of GPUs.

AI Co-coding
Ready-to-use GPUs
Best Pricing
Get Started

Hyper Newsletters

Subscribe to our latest updates
We will deliver the latest updates of the week to your inbox at nine o'clock every Monday morning
Powered by MailChimp
Go-ICP: A Globally Optimal Solution to 3D ICP Point-Set Registration | Papers | HyperAI