IP Library Granted Patent US 12,731,271
Granted Patent B2
US 12,731,271 · App. 18/425,436 · Granted Sep 8, 2026

Active learning system and method

Inventors: Edwardo Martinez (Fremont, CA); Vitor Campagnolo Guizilini (Santa Clara, CA); Erin A McColl (Los Gatos, CA)
Assignees: Toyota Research Institute, Inc.; Toyota Jidosha Kabushiki Kaisha
G06T7/30G06N3/091G06T7/50G06T2207/10028G06T2207/20081G06T2207/30252
View Patent ↗
Loading inventors, assignments & file history…
Monitor This Case
Get email alerts when status or documents change.
Order Certified Copies
Most orders are placed with the USPTO same day — all within 24 business hours.
Order via The Patent Place →
Pre-filled with this patent's details
Quick Facts
Patent No.
US 12,731,271
App. No.
18/425,436
Granted
Sep 8, 2026
Kind
B2
Abstract

Disclosed are systems and methods for training an active learning system. In one example, a method for training an active learning system includes the steps of generating a point cloud based on a first image captured at a first pose, projecting the point cloud into a second image captured at a second pose to determine an overlap ratio between the point cloud and the second image, and using the second image to train the active learning system based on probabilities associated with the points of the point cloud projected into the second image and the overlap ratio.

Claims (43)

1 . An active learning system comprising:

a processor; and

a memory in communication with the processor, the memory including a training module with instructions that, when executed by the processor, cause the processor to:

generate a point cloud based on a first image captured at a first pose, wherein the point cloud comprises a plurality of points having probabilities associated with the points, the probabilities indicating a confidence that the points are correctly providing depth,

project the point cloud into a second image comprising an adjacent frame with respect to the first image captured at a second pose to determine an overlap ratio between the point cloud and the second image, and

use the second image to train the active learning system in a self-supervised fashion when (i) an overall confidence of the point cloud determined from the probabilities satisfies a confidence criterion and (ii) the overlap ratio satisfies an overlap criterion.

2 . The active learning system of claim 1 , wherein the training module further includes instructions that, when executed by the processor, cause the processor to use the second image to train the active learning system when the overlap ratio is above a lower threshold and below an upper threshold.

3 . The active learning system of claim 1 , wherein the overlap ratio is a percent of pixels of the second image that the point cloud projects onto.

4 . The active learning system of claim 1 , wherein the training module further includes instructions that, when executed by the processor, cause the processor to use the second image to train the active learning system when the probabilities associated with the points of the point cloud projected into the second image are below a probability threshold.

5 . The active learning system of claim 1 , wherein the training module further includes instructions that, when executed by the processor, cause the processor to move the active learning system to a third pose based on at least one of the probabilities associated with the points of the point cloud projected into the second image and the overlap ratio.

6 . The active learning system of claim 5 , wherein the training module further includes instructions that, when executed by the processor, cause the processor to:

not use the second image to train the active learning system when the overlap ratio is below a lower threshold or above an upper threshold, and

move the active learning system to the third pose.

7 . The active learning system of claim 5 , wherein the training module further includes instructions that, when executed by the processor, cause the processor to:

not use the second image to train the active learning system when the probabilities associated with the points of the point cloud projected into the second image are above a probability threshold and

move the active learning system to the third pose.

8 . A method for training an active learning system comprising steps of:

generating a point cloud based on a first image captured at a first pose, wherein the point cloud comprises a plurality of points having probabilities associated with the points, the probabilities indicating a confidence that the points are correctly providing depth;

projecting the point cloud into a second image comprising an adjacent frame with respect to the first image captured at a second pose to determine an overlap ratio between the point cloud and the second image; and

using the second image to train the active learning system in a self-supervised fashion when (i) an overall confidence of the point cloud determined from the probabilities satisfies a confidence criterion and (ii) the overlap ratio satisfies an overlap criterion.

9 . The method of claim 8 , further comprising the step of using the second image to train the active learning system when the overlap ratio is above a lower threshold and below an upper threshold.

10 . The method of claim 8 , wherein the overlap ratio is a percent of pixels of the second image that the point cloud projects onto.

11 . The method of claim 8 , further comprising the step of using the second image to train the active learning system when the probabilities associated with the points of the point cloud projected into the second image are below a probability threshold.

12 . The method of claim 8 , further comprising the step of moving the active learning system to a third pose based on at least one of the probabilities associated with the points of the point cloud projected into the second image and the overlap ratio.

13 . The method of claim 12 , further comprising the steps of:

not using the second image to train the active learning system when the overlap ratio is below a lower threshold or above an upper threshold; and

moving the active learning system to the third pose.

14 . The method of claim 12 , further comprising the steps of:

not using the second image to train the active learning system when the probabilities associated with the points of the point cloud projected into the second image are above a probability threshold; and

moving the active learning system to the third pose.

15 . A non-transitory computer-readable medium including instructions for training an active learning system, the instructions, when executed by a processor, cause the processor to:

generate a point cloud based on a first image captured at a first pose, wherein the point cloud comprises a plurality of points having probabilities associated with the points, the probabilities indicating a confidence that the points are correctly providing depth;

project the point cloud into a second image comprising an adjacent frame with respect to the first image captured at a second pose to determine an overlap ratio between the point cloud and the second image; and

use the second image to train the active learning system in a self-supervised fashion when (i) an overall confidence of the point cloud determined from the probabilities satisfies a confidence criterion and (ii) the overlap ratio satisfies an overlap criterion.

16 . The non-transitory computer-readable medium of claim 15 , further including instructions that, when executed by the processor, cause the processor to use the second image to train the active learning system when the overlap ratio is above a lower threshold and below an upper threshold.

17 . The non-transitory computer-readable medium of claim 15 , further including instructions that, when executed by the processor, cause the processor to use the second image to train the active learning system when the probabilities associated with the points of the point cloud projected into the second image are below a probability threshold.

18 . The non-transitory computer-readable medium of claim 15 , further including instructions that, when executed by the processor, cause the processor to move the active learning system to a third pose based on at least one of the probabilities associated with the points of the point cloud projected into the second image and the overlap ratio.

19 . The non-transitory computer-readable medium of claim 18 , further including instructions that, when executed by the processor, cause the processor to:

not use the second image to train the active learning system when the overlap ratio is below a lower threshold or above an upper threshold; and

move the active learning system to the third pose.

20 . The non-transitory computer-readable medium of claim 18 , further including instructions that, when executed by the processor, cause the processor to:

not use the second image to train the active learning system when the probabilities associated with the points of the point cloud projected into the second image are above a probability threshold; and

move the active learning system to the third pose.

Assignments (1)
ASSIGNMENT OF ASSIGNOR'S INTEREST Recorded Feb 12, 2024
From: MARTINEZ, EDWARDO; GUIZILINI, VITOR CAMPAGNOLO; MCCOLL, ERIN A.
To: TOYOTA RESEARCH INSTITUTE, INC.; TOYOTA JIDOSHA KABUSHIKI KAISHA
Reel/Frame 066435/0625 →
Continuity (1)
Related Publication 20250245842A1 · Jul 31, 2025
References Cited (15)
US 10976421B2 · Zhang et al. · 2021 [cited by applicant]
US 20240371035A1 · Jadhav · 2024 [cited by examiner]
CN 113281780A · 2021 [cited by examiner]
CN 115962773A · 2023 [cited by applicant]
CN 116758166A · 2023 [cited by examiner]
Blok, P. M., Kootstra, G., Elghor, H. E., Diallo, B., van Evert, F. K., & van Henten, E. J. (2022). Active learning with MaskAL reduces annotation effort for training Mask R-CNN on a broccoli dataset with visually simil… [cited by examiner]
Lin, Y., Vosselman, G., Cao, Y., & Yang, M. Y. (2020). Active and incremental learning for semantic ALS point cloud segmentation. ISPRS journal of photogrammetry and remote sensing, 169, 73-92. (Year: 2020). [cited by examiner]
Rathnayake, D., Radhakrishnan, M., Hwang, I., & Misra, A. (May 2023). LILOC: Enabling precise 3D localization in dynamic indoor environments using LiDARs. In Proceedings of the 8th ACM/IEEE Conference on Internet of Thi… [cited by examiner]
Zeng, W., Karaoglu, S., & Gevers, T. (2018). Inferring point clouds from single monocular images by depth intermediation. arXiv preprint arXiv:1812.01402. (Year: 2018). [cited by examiner]
Wang et al., “Deep Closest Point: Learning Representations for Point Cloud Registration,” Proceedings of the IEEE/CVF international conference on computer vision, 2019, pp. 3523-3532. [cited by applicant]
Chen et al., “PLADE: A Plane-based Descriptor for Point Cloud Registration with Small Overlap,” IEEE Transactions on Geoscience and Remote Sensing, 2019, pp. 2530-2540. [cited by applicant]
Blanton et al., “A Structure-Aware Method for Direct Pose Estimation,” Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, 2022, pp. 2019-2028. [cited by applicant]
Trevor et al., “Efficient Organized Point Cloud Segmentation with Connected Components,” Semantic Perception Mapping and Exploration (SPME), 2013, pp. 251-257. [cited by applicant]
Mei et al., “Overlap-guided Gaussian Mixture Models for Point Cloud Registration,” Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, 2023, pp. 4511-4520. [cited by applicant]
Zhang et al., “Monocular weakly supervised depth and pose estimation method based on multi-information fusion,” Kuwait Journal of Science 49.3, 2022, 23 pages. [cited by applicant]