1
|
Mukhandi H, Ferreira JF, Peixoto P. SyS3DS: Systematic Sampling of Large-Scale LiDAR Point Clouds for Semantic Segmentation in Forestry Robotics. SENSORS (BASEL, SWITZERLAND) 2024; 24:823. [PMID: 38339539 PMCID: PMC10856877 DOI: 10.3390/s24030823] [Citation(s) in RCA: 0] [Impact Index Per Article: 0] [Reference Citation Analysis] [Abstract] [Key Words] [Track Full Text] [Subscribe] [Scholar Register] [Received: 11/02/2023] [Revised: 01/09/2024] [Accepted: 01/23/2024] [Indexed: 02/12/2024]
Abstract
Recently, new semantic segmentation and object detection methods have been proposed for the direct processing of three-dimensional (3D) LiDAR sensor point clouds. LiDAR can produce highly accurate and detailed 3D maps of natural and man-made environments and is used for sensing in many contexts due to its ability to capture more information, its robustness to dynamic changes in the environment compared to an RGB camera, and its cost, which has decreased in recent years and which is an important factor for many application scenarios. The challenge with high-resolution 3D LiDAR sensors is that they can output large amounts of 3D data with up to a few million points per second, which is difficult to process in real time when applying complex algorithms and models for efficient semantic segmentation. Most existing approaches are either only suitable for relatively small point clouds or rely on computationally intensive sampling techniques to reduce their size. As a result, most of these methods do not work in real time in realistic field robotics application scenarios, making them unsuitable for practical applications. Systematic point selection is a possible solution to reduce the amount of data to be processed. Although our approach is memory and computationally efficient, it selects only a small subset of points, which may result in important features being missed. To address this problem, our proposed systematic sampling method called SyS3DS (Systematic Sampling for 3D Semantic Segmentation) incorporates a technique in which the local neighbours of each point are retained to preserve geometric details. SyS3DS is based on the graph colouring algorithm and ensures that the selected points are non-adjacent in order to obtain a subset of points that are representative of the 3D points in the scene. To take advantage of the ensemble learning method, we pass a different subset of nodes for each epoch. This leverages a new technique called auto-ensemble, where ensemble learning is proposed as a collection of different learning models instead of tuning different hyperparameters individually during training and validation. SyS3DS has been shown to process up to 1 million points in a single pass. It outperforms the state of the art in efficient semantic segmentation on large datasets such as Semantic3D. We also present a preliminary study on the validity of the performance of LiDAR-only data, i.e., intensity values from LiDAR sensors without RGB values for semi-autonomous robot perception.
Collapse
Affiliation(s)
- Habibu Mukhandi
- Institute of Systems and Robotics, University of Coimbra, 3030-290 Coimbra, Portugal
| | - Joao Filipe Ferreira
- Institute of Systems and Robotics, University of Coimbra, 3030-290 Coimbra, Portugal
- Computational Intelligence and Applications Research Group, Department of Computer Science, School of Science and Technology, Nottingham NG11 8NS, UK
| | - Paulo Peixoto
- Institute of Systems and Robotics, University of Coimbra, 3030-290 Coimbra, Portugal
- University of Coimbra, Department of Electrical and Computer Engineering, 3030-290 Coimbra, Portugal
| |
Collapse
|
2
|
Shin C, Hong SH, Jeong H, Yoon H, Koh B. All-in-one encoder/decoder approach for non-destructive identification of 3D-printed objects. MATHEMATICAL BIOSCIENCES AND ENGINEERING : MBE 2022; 19:14102-14115. [PMID: 36654082 DOI: 10.3934/mbe.2022657] [Citation(s) in RCA: 0] [Impact Index Per Article: 0] [Reference Citation Analysis] [Abstract] [Key Words] [MESH Headings] [Track Full Text] [Subscribe] [Scholar Register] [Indexed: 06/17/2023]
Abstract
This paper presents an all-in-one encoder/decoder approach for the nondestructive identification of three-dimensional (3D)-printed objects. The proposed method consists of three parts: 3D code insertion, terahertz (THz)-based detection, and code extraction. During code insertion, a relevant one-dimensional (1D) identification code is generated to identify the 3D-printed object. A 3D barcode corresponding to the identification barcode is then generated and inserted into a blank bottom area inside the object's stereolithography (STL) file. For this objective, it is necessary to find an appropriate area of the STL file and to merge the 3D barcode and the model within the STL file. Next the information generated inside the object is extracted by using THz waves that are transmitted and reflected by the output 3D object. Finally, the resulting THz signal from the target object is detected and analyzed to extract the identification information. We implemented and tested the proposed method using a 3D graphic environment and a THz time-domain spectroscopy system. The experimental results indicate that one-dimensional barcodes are useful for identifying 3D-printed objects because they are simple and practical to process. Furthermore, information efficiency can be increased by using an integral fast Fourier transform to identify any code located in areas deeper within the object. As 3D printing is used in various fields, the proposed method is expected to contribute to the acceleration of the distribution of 3D printing empowered by the integration of the internal code insertion and recognition process.
Collapse
Affiliation(s)
- Choonsung Shin
- Graduate School of Culture, Chonnam National University, Gwangju, Korea
| | - Sung-Hee Hong
- Hologram Research Center, Korea Electronics Technology Institute, Seoul, Korea
| | - Hieyoung Jeong
- Department of Artificial Intelligence Convergence, Chonnam National University, Gwangju, Korea
| | - Hyoseok Yoon
- Division of Computer Engineering, Hanshin University, Osan-si, Gyeonggi-do, Korea
| | - Byoungsoo Koh
- CT R & D Team, Korea Creative Content Agency, Naju-si, Jeollanam-do, Korea
| |
Collapse
|
3
|
Abstract
Accurate semantic analysis of LiDAR point clouds enables the interaction between intelligent vehicles and the real environment. This paper proposes a hybrid 2D and 3D Hough Net by combining 3D global Hough features and 2D local Hough features with a classification deep learning network. Firstly, the 3D object point clouds are mapped into the 3D Hough space to extract the global Hough features. The generated global Hough features are input into the 3D convolutional neural network for training global features. Furthermore, a multi-scale critical point sampling method is designed to extract critical points in the 2D views projected from the point clouds to reduce the computation of redundant points. To extract local features, a grid-based dynamic nearest neighbors algorithm is designed by searching the neighbors of the critical points. Finally, the two networks are connected to the full connection layer, which is input into fully connected layers for object classification.
Collapse
|
4
|
Hong S, Park D. Runtime ML-DL Hybrid Inference Platform Based on Multiplexing Adaptive Space-Time Resolution for Fast Car Incident Prevention in Low-Power Embedded Systems. SENSORS 2022; 22:s22082998. [PMID: 35458983 PMCID: PMC9024881 DOI: 10.3390/s22082998] [Citation(s) in RCA: 0] [Impact Index Per Article: 0] [Reference Citation Analysis] [Abstract] [Key Words] [MESH Headings] [Track Full Text] [Download PDF] [Figures] [Subscribe] [Scholar Register] [Received: 03/09/2022] [Revised: 04/04/2022] [Accepted: 04/08/2022] [Indexed: 02/04/2023]
Abstract
Forward vehicle detection is the key technique to preventing car incidents in front. Artificial intelligence (AI) techniques are used to more accurately detect vehicles, but AI-based vehicle detection takes a lot of processing time due to its high computational complexity. When there is a risk of collision with a vehicle in front, the slow detection speed of the vehicle may lead to an accident. To quickly detect a vehicle in real-time, a high-speed and lightweight vehicle detection technique with similar detection performance to that of an existing AI-based vehicle detection is required. In addition, to apply forward collision warning system (FCWS) technology to vehicles, it is important to provide high performance based on low-power embedded systems because the vehicle’s battery consumption must remain low. The vehicle detection algorithm occupies the most resources in FCWS. To reduce power consumption, it is important to reduce the computational complexity of an algorithm, that is, the amount of resources required to run it. This paper describes a method for fast, accurate forward vehicle detection using machine learning and deep learning. To detect a vehicle in consecutive images consistently, a Kalman filter is used to predict the bounding box based on the tracking algorithm and correct it based on the detection algorithm. As a result, its vehicle detection speed is about 25.85 times faster than deep-learning-based object detection is, and its detection accuracy is better than machine-learning-based object detection is.
Collapse
|
5
|
A Convolution Neural Network-Based Representative Spatio-Temporal Documents Classification for Big Text Data. APPLIED SCIENCES-BASEL 2022. [DOI: 10.3390/app12083843] [Citation(s) in RCA: 1] [Impact Index Per Article: 0.5] [Reference Citation Analysis] [Abstract] [Track Full Text] [Subscribe] [Scholar Register] [Indexed: 02/04/2023]
Abstract
With the proliferation of mobile devices, the amount of social media users and online news articles are rapidly increasing, and text information online is accumulating as big data. As spatio-temporal information becomes more important, research on extracting spatiotemporal information from online text data and utilizing it for event analysis is being actively conducted. However, if spatiotemporal information that does not describe the core subject of a document is extracted, it is rather difficult to guarantee the accuracy of core event analysis. Therefore, it is important to extract spatiotemporal information that describes the core topic of a document. In this study, spatio-temporal information describing the core topic of a document is defined as ‘representative spatio-temporal information’, and documents containing representative spatiotemporal information are defined as ‘representative spatio-temporal documents’. We proposed a character-level Convolution Neuron Network (CNN)-based document classifier to classify representative spatio-temporal documents. To train the proposed CNN model, 7400 training data were constructed for representative spatio-temporal documents. The experimental results show that the proposed CNN model outperforms traditional machine learning classifiers and existing CNN-based classifiers.
Collapse
|
6
|
A Supervoxel-Based Random Forest Method for Robust and Effective Airborne LiDAR Point Cloud Classification. REMOTE SENSING 2022. [DOI: 10.3390/rs14061516] [Citation(s) in RCA: 4] [Impact Index Per Article: 2.0] [Reference Citation Analysis] [Abstract] [Track Full Text] [Subscribe] [Scholar Register] [Indexed: 12/04/2022]
Abstract
As an essential part of point cloud processing, autonomous classification is conventionally used in various multifaceted scenes and non-regular point distributions. State-of-the-art point cloud classification methods mostly process raw point clouds, using a single point as the basic unit and calculating point cloud features by searching local neighbors via the k-neighborhood method. Such methods tend to be computationally inefficient and have difficulty obtaining accurate feature descriptions due to inappropriate neighborhood selection. In this paper, we propose a robust and effective point cloud classification approach that integrates point cloud supervoxels and their locally convex connected patches into a random forest classifier, which effectively improves the point cloud feature calculation accuracy and reduces the computational cost. Considering the different types of point cloud feature descriptions, we divide features into three categories (point-based, eigen-based, and grid-based) and accordingly design three distinct feature calculation strategies to improve feature reliability. Two International Society of Photogrammetry and Remote Sensing benchmark tests show that the proposed method achieves state-of-the-art performance, with average F1-scores of 89.16 and 83.58, respectively. The successful classification of point clouds with great variation in elevation also demonstrates the reliability of the proposed method in challenging scenes.
Collapse
|
7
|
Bae C, Lee YC, Yu W, Lee S. Spherically Stratified Point Projection: Feature Image Generation for Object Classification Using 3D LiDAR Data. SENSORS 2021; 21:s21237860. [PMID: 34883864 PMCID: PMC8659660 DOI: 10.3390/s21237860] [Citation(s) in RCA: 0] [Impact Index Per Article: 0] [Reference Citation Analysis] [Abstract] [Key Words] [Track Full Text] [Download PDF] [Figures] [Subscribe] [Scholar Register] [Received: 10/05/2021] [Revised: 11/15/2021] [Accepted: 11/16/2021] [Indexed: 11/16/2022]
Abstract
Three-dimensional point clouds have been utilized and studied for the classification of objects at the environmental level. While most existing studies, such as those in the field of computer vision, have detected object type from the perspective of sensors, this study developed a specialized strategy for object classification using LiDAR data points on the surface of the object. We propose a method for generating a spherically stratified point projection (sP2) feature image that can be applied to existing image-classification networks by performing pointwise classification based on a 3D point cloud using only LiDAR sensors data. The sP2’s main engine performs image generation through spherical stratification, evidence collection, and channel integration. Spherical stratification categorizes neighboring points into three layers according to distance ranges. Evidence collection calculates the occupancy probability based on Bayes’ rule to project 3D points onto a two-dimensional surface corresponding to each stratified layer. Channel integration generates sP2 RGB images with three evidence values representing short, medium, and long distances. Finally, the sP2 images are used as a trainable source for classifying the points into predefined semantic labels. Experimental results indicated the effectiveness of the proposed sP2 in classifying feature images generated using the LeNet architecture.
Collapse
Affiliation(s)
- Chulhee Bae
- Department of Mechanical Engineering, Kongju National University, Cheonan 31080, Korea;
| | - Yu-Cheol Lee
- Artificial Intelligence Laboratory, ETRI, Daejeon 34129, Korea; (Y.-C.L.); (W.Y.)
- Department of Computer Software, University of Science and Technology, Daejeon 34113, Korea
| | - Wonpil Yu
- Artificial Intelligence Laboratory, ETRI, Daejeon 34129, Korea; (Y.-C.L.); (W.Y.)
| | - Sejin Lee
- Department of Mechanical Engineering, Kongju National University, Cheonan 31080, Korea;
- Correspondence:
| |
Collapse
|
8
|
Feature extraction and machine learning techniques for identifying historic urban environmental hazards: New methods to locate lost fossil fuel infrastructure in US cities. PLoS One 2021; 16:e0255507. [PMID: 34347840 PMCID: PMC8336811 DOI: 10.1371/journal.pone.0255507] [Citation(s) in RCA: 2] [Impact Index Per Article: 0.7] [Reference Citation Analysis] [Abstract] [Track Full Text] [Download PDF] [Figures] [Journal Information] [Subscribe] [Scholar Register] [Received: 02/25/2021] [Accepted: 07/18/2021] [Indexed: 11/19/2022] Open
Abstract
U.S. cities contain unknown numbers of undocumented "manufactured gas" sites, legacies of an industry that dominated energy production during the late-19th and early-20th centuries. While many of these unidentified sites likely contain significant levels of highly toxic and biologically persistent contamination, locating them remains a significant challenge. We propose a new method to identify manufactured gas production, storage, and distribution infrastructure in bulk by applying feature extraction and machine learning techniques to digitized historic Sanborn fire insurance maps. Our approach, which relies on a two-part neural network to classify candidate map regions, increases the rate of site identification 20-fold compared to unaided visual coding.
Collapse
|
9
|
Object Classification with Roadside LiDAR Data Using a Probabilistic Neural Network. ELECTRONICS 2021. [DOI: 10.3390/electronics10070803] [Citation(s) in RCA: 2] [Impact Index Per Article: 0.7] [Reference Citation Analysis] [Abstract] [Track Full Text] [Subscribe] [Scholar Register] [Indexed: 11/16/2022]
Abstract
Object classification is important information for different transportation areas. This research developed a probabilistic neural network (PNN) classifier for object classification using roadside Light Detection and Ranging (LiDAR). The objective was to classify the road user on the urban road into one of four classes: Pedestrian, bicycle, passenger car, and truck. Five features calculated from the point cloud generated from the roadside LiDAR were selected to represent the difference between different classes. A total of 2736 records (2062 records for training, and 674 records for testing) were manually marked for training and testing the PNN algorithm. The data were collected at three different sites representing different scenarios. The performance of the classification was evaluated by comparing the result of the PNN with those of the support vector machine (SVM) and the random forest (RF). The comparison results showed that the PNN can provide the results of classification with the highest accuracy among the three investigated methods. The overall accuracy of the PNN for object classification was 97.6% using the testing database. The errors in the classification results were also diagnosed. Discussions about the direction of future studies were also provided at the end of this paper.
Collapse
|