Machine visual perception from sim-to-real transfer learning for autonomous docking maneuvers DOI Creative Commons
Derek Worth, Jeffrey Choate, Ryan Raettig

et al.

Neural Computing and Applications, Journal Year: 2024, Volume and Issue: unknown

Published: Dec. 3, 2024

Abstract This paper presents a comprehensive approach to enhancing autonomous docking maneuvers through machine visual perception and sim-to-real transfer learning. By leveraging relative vectoring techniques, we aim replicate the human ability execute precise operations. Our study focuses on aerial refueling as use case, demonstrating significant advancements in navigation object detection. We introduce novel method for aligning digital twins using fiducial targets motion capture data, which facilitates accurate pose estimation from real-world imagery. Additionally, develop cost-efficient annotation automation techniques generating high-quality You Only Look Once training data. Experimental results indicate that our learning methodologies enable reliable conditions, achieving error margins of less than 3 cm at contact (when vehicles are approximately 4 m camera) maintaining performance over 56 fps. The research findings underscore potential augmented reality scene augmentation improving model generalization performance, bridging gap between simulation applications. work lays groundwork deploying systems complex dynamic environments, minimizing intervention operational efficiency.

Language: Английский

From Simulation to Reality: Transfer Learning for Automating Pseudo‐Labeling of Real and Infrared Imagery DOI Creative Commons
Jeffrey Choate, Derek Worth, Scott Nykl

et al.

Advanced Intelligent Systems, Journal Year: 2025, Volume and Issue: unknown

Published: Feb. 2, 2025

Training a convolutional neural network (CNN) for real‐world applications is challenging due to the requirement of high‐quality labeled imagery. This study employs pseudo‐labeling and transfer learning, built upon 6D pose estimation framework. A CNN trained on synthetic images predicts bounding boxes (bbox) an object's components in real image. With as few four bbox predictions, framework solves relative camera reprojects bboxes all onto that The reprojections allow filtering bad common issue pseudo‐labeling. Thereby, enabling automated labeling large datasets with minimal human intervention. Tested color long‐wave infrared imagery captured during December 2023 flight tests, this process demonstrates increased enhanced performance across situations, reduced reprojection error, stabilized predictions. technique significant it enables without expensive truth systems, requiring only camera. It supports learning previously known calibrations, facilitating data creation impractical‐to‐simulate sensors. Ultimately, approach provides low‐cost precise method creating CNNs operationally relevant data, unattainable by everyday user.

Language: Английский

Citations

0
KSL-POSE: A Real-Time 2D Human Pose Estimation Method Based on Modified YOLOv8-Pose Framework DOI Creative Commons

Tianyi Lu,

Ke Cheng, Xuecheng Hua

et al.

Sensors, Journal Year: 2024, Volume and Issue: 24(19), P. 6249 - 6249

Published: Sept. 26, 2024

Two-dimensional human pose estimation aims to equip computers with the ability accurately recognize keypoints and comprehend their spatial contexts within media content. However, accuracy of real-time diminishes when processing images occluded body parts or overlapped individuals. To address these issues, we propose a method based on YOLO framework. We integrate convolutional concepts Kolmogorov–Arnold Networks (KANs) through introducing non-linear activation functions enhance feature extraction capabilities kernels. Moreover, improve detection small target keypoints, cross-stage partial (CSP) approach utilize object pyramid (SOEP) module for integration. also innovatively incorporate layered shared convolution batch normalization head (LSCB), consisting multiple layers layers, enable fusion low utilization model parameters. Given structure purpose proposed model, name it KSL-POSE. Compared baseline YOLOv8l-POSE, KSL-POSE achieves significant improvements, increasing average by 1.5% public MS COCO 2017 data set. Furthermore, demonstrates competitive performance CrowdPOSE set, thus validating its generalization ability.

Language: Английский

Citations

1
CBA-YOLOv5s: A hip dysplasia detection algorithm based on YOLOv5s using angle consistency and bi-level routing attention DOI

Jia Lv,

Junliang Che,

Xin Chen

et al.

Biomedical Signal Processing and Control, Journal Year: 2024, Volume and Issue: 95, P. 106482 - 106482

Published: May 23, 2024

Language: Английский

Citations

0
A Robust Pointer Meter Reading Recognition Method Based on TransUNet and Perspective Transformation Correction DOI Open Access

Liufan Tan,

Wanneng Wu,

Jinxin Ding

et al.

Electronics, Journal Year: 2024, Volume and Issue: 13(13), P. 2436 - 2436

Published: June 21, 2024

The automatic reading recognition of pointer meters plays a crucial role in data monitoring and analysis intelligent substations. Existing meter methods struggle to address challenging difficulties such as image distortion varying illumination. To enhance their robustness accuracy, this study proposes novel approach that leverages the TransUNet semantic segmentation model perspective transformation correction method. Initially, dial is localized from natural background using YOLOv8. Subsequently, after enhancing with Gamma technology, scale lines within are extracted model. distorted or rotated can then be corrected through transformation. Finally, readings accurately obtained by Weighted Angle Method (WAM). Ablative comparative experiments on two self-collected datasets clearly verify effectiveness proposed method, accuracy 97.81% Simple-MeterData 93.39% Complex-MeterData, respectively.

Language: Английский

Citations

0
Leveraging Advanced Computer Vision for Hazardous Behavior Monitoring on Campus Safety Maintenance DOI

Shang-Te Tsai,

Zong-Rong Wu,

Yu‐Cheng Chang

et al.

Published: April 19, 2024

Language: Английский

Citations

0
Machine visual perception from sim-to-real transfer learning for autonomous docking maneuvers DOI Creative Commons
Derek Worth, Jeffrey Choate, Ryan Raettig

et al.

Neural Computing and Applications, Journal Year: 2024, Volume and Issue: unknown

Published: Dec. 3, 2024

Abstract This paper presents a comprehensive approach to enhancing autonomous docking maneuvers through machine visual perception and sim-to-real transfer learning. By leveraging relative vectoring techniques, we aim replicate the human ability execute precise operations. Our study focuses on aerial refueling as use case, demonstrating significant advancements in navigation object detection. We introduce novel method for aligning digital twins using fiducial targets motion capture data, which facilitates accurate pose estimation from real-world imagery. Additionally, develop cost-efficient annotation automation techniques generating high-quality You Only Look Once training data. Experimental results indicate that our learning methodologies enable reliable conditions, achieving error margins of less than 3 cm at contact (when vehicles are approximately 4 m camera) maintaining performance over 56 fps. The research findings underscore potential augmented reality scene augmentation improving model generalization performance, bridging gap between simulation applications. work lays groundwork deploying systems complex dynamic environments, minimizing intervention operational efficiency.

Language: Английский

Citations

0