Ma, H, Xiong, R, Wang, Y, Kodagoda, S & Shi, L 2018, 'Towards open-set semantic labeling in 3D point clouds : Analysis on the unknown class', Neurocomputing, vol. 275, pp. 1282-1294.View/Download from: Publisher's site
© 2017 Elsevier B.V. There has been a growing interest in the research of semantic labeling on scenes represented by 3D point clouds. A fundamental issue that has been largely ignored is the unavoidable presence of unknown objects and the lack of effective ways of dealing with them. Traditional methods usually label unknown objects as one of the pre-trained classes which is either a meaningful target class or a defined unknown class that collectively refers to all uninterested objects. Due to the fact that the class of unknown in essence is a collection of many unseen or uninterested classes, in which the in-class variation is significant and less manageable. It is challenging to solve the unknown problem in a pre-trained manner. In order to advance the research on semantic labeling with the presence of unknown objects, this study investigates the feasibility of adopting an open-set approach, i.e. train a model without unknown objects and reject them accurately in the test. In this paper, we propose a method that exploits the conflict of different labeling results in order to withstand the negative effect of unknown objects. The proposed framework relies on a Conditional Random Field (CRF) to capture inherent spatial relationships and appearance similarities between objects, and employs a Probability of Inclusion Support Vector Machine (P I SVM) to estimate an unknown likelihood for each training class. The probabilistic outputs from both CRF and P I SVM are then proposed to be combined under the Dempster Shafer theory for conflict measurement and unkno wn rejection. The novelty lies in that the method encodes both contextual constrains and unknown likelihood for performance enhancement. Comprehensive experimental results on publicly available data sets are presented to show the negative effects of unknown objects and the improvements on labeling accuracy achieved by the proposed method.
Miro, JV, Ulapane, N, Shi, L, Hunt, D & Behrens, M 2018, 'Robotic pipeline wall thickness evaluation for dense nondestructive testing inspection', JOURNAL OF FIELD ROBOTICS, vol. 35, no. 8, pp. 1293-1310.View/Download from: Publisher's site
Shi, L & Miro, JV 2018, 'Towards optimised and reconstructable sampling inspection of pipe integrity for improved efficiency of non-destructive testing', Water Science and Technology: Water Supply, vol. 18, no. 2, pp. 515-523.View/Download from: Publisher's site
© IWA Publishing 2018. This work proposes a sampling inspection framework for point measurement non-destructive testing of pipelines to improve its time and cost efficiencies. Remaining pipe wall thickness data from limited dense inspection are modelled with spatial statistics approaches. The spatial dependence in the available data and some subjective requirements provide a reference for selecting a most efficient sampling inspection scheme. With the learned model and the selected sampling scheme, the effort of inspecting the residual part of the same pipeline or cohort will be significantly reduced from dense inspection to sampling inspection, and the full information can be reconstructed from samples while maintaining a reasonable accuracy. The recovered thickness map can be used as an equivalent measure to the dense inspection for subsequent structural analysis for failure risk estimation or remaining life assessment.
Wang, M, Liu, Y, Su, D, Liao, Y, Shi, L, Xu, J & Valls Miro, J 2018, 'Accurate and Real-Time 3-D Tracking for the Following Robots by Fusing Vision and Ultrasonar Information', IEEE/ASME Transactions on Mechatronics, vol. 23, no. 3, pp. 997-1006.View/Download from: Publisher's site
© 1996-2012 IEEE. Acquiring the accurate three-dimensional (3-D) position of a target person around a robot provides valuable information that is applicable to a wide range of robotic tasks, especially for promoting the intelligent manufacturing processes of industries. This paper presents a real-time robotic 3-D human tracking system that combines a monocular camera with an ultrasonic sensor by an extended Kalman filter (EKF). The proposed system consists of three submodules: a monocular camera sensor tracking module, an ultrasonic sensor tracking module, and the multisensor fusion algorithm. An improved visual tracking algorithm is presented to provide 2-D partial location estimation. The algorithm is designed to overcome severe occlusions, scale variation, target missing, and achieve robust redetection. The scale accuracy is further enhanced by the estimated 3-D information. An ultrasonic sensor array is employed to provide the range information from the target person to the robot, and time of flight is used for the 2-D partial location estimation. EKF is adopted to sequentially process multiple, heterogeneous measurements arriving in an asynchronous order from the vision sensor, and the ultrasonic sensor separately. In the experiments, the proposed tracking system is tested in both a simulation platform and actual mobile robot for various indoor and outdoor scenes. The experimental results show the persuasive performance of the 3-D tracking system in terms of both the accuracy and robustness.
Wang, S, Kodagoda, S, Shi, L & Dai, X 2018, 'Two-Stage Road Terrain Identification Approach for Land Vehicles Using Feature-Based and Markov Random Field Algorithm', IEEE Intelligent Systems, vol. 33, no. 1, pp. 29-39.View/Download from: Publisher's site
© 2001-2011 IEEE. Road terrain identification is one of the important tasks for driving assistant systems or autonomous land vehicles. It plays a key role in improving driving strategy and enhancing fuel efficiency. In this paper, a two-stage approach using multiple sensors is presented. In the first stage, a feature-based identification approach is performed using an accelerometer, a camera, and downward-looking and forward-looking laser range finders (LRFs). This produces four classification label sequences. In the second stage, a majority vote is implemented for each label sequences to match them into synchronized road patches. Then a Markov Random Field (MRF) model is designed to generate the final optimized identification results to improve the forward-looking LRF. This approach enables the vehicle to observe the upcoming road terrain before moving onto it by fusing all the classification results using an MRF algorithm. The experiments show this approach improved the terrain identification accuracy and robustness significantly for some familiar road terrains.
Liao, Y, Kodagoda, S, Wang, Y, Shi, L & Liu, Y 2017, 'Place Classification With a Graph Regularized Deep Neural Network', IEEE Transactions on Cognitive and Developmental Systems, vol. 9, no. 4, pp. 304-315.View/Download from: Publisher's site
Wang, S, Kodagoda, S, Shi, L & Wang, H 2017, 'Road-Terrain Classification for Land Vehicles: Employing an Acceleration-Based Approach', IEEE Vehicular Technology Magazine, vol. 12, no. 3, pp. 34-41.View/Download from: Publisher's site
© 2017 IEEE. The perception of the environment around a land vehicle plays a crucial role for its driving assistant system. Knowledge of the road terrain is useful for handling its characteristics while driving the vehicles and improving passengers' safety and comfort. In this article, an approach to classifying road-terrain vehicles is presented. An accelerometer is mounted on the suspension of the vehicle to measure the vibration that represents the characteristics of the road terrain, and the road profile can be calculated by knowing the speed and one-quarter-dynamic model of the vehicle. The optimized classifier and features, speed independency, and the effect of employing principal component analysis (PCA) are investigated, and the simulation shows that this acceleration-based approach is feasible for land vehicles in a range of outdoor scenarios.
Wang, S, Kodagoda, S, Shi, L & Xu, N 2017, 'Lidar-based road terrain recognition for passenger vehicles', International Journal of Vehicle Design, vol. 74, no. 2, pp. 153-165.View/Download from: Publisher's site
Copyright © 2017 Inderscience Enterprises Ltd. The road terrain type is important information about a passenger vehicle's surroundings. It suggests an appropriate control algorithm and driving strategy. In this paper, a Lidar sensor is employed to reconstruct the road surface and extract features for terrain classification. The experiment vehicle was driven on four specific road terrains at a variety of speeds. The speed dependency and the effect of using principal component analysis were investigated. The simulation experimental results show that this Lidar sensor-based approach is feasible and robust for passenger vehicles in a range of outdoor scenarios.
Shi, L & Kodagoda, S 2013, 'Towards generalization of semi-supervised place classification over generalized Voronoi graph', Robotics And Autonomous Systems, vol. 61, no. 8, pp. 785-796.View/Download from: Publisher's site
With the progress of humanrobot interaction (HRI), the ability of a robot to perform high-level tasks in complex environments is fast becoming an essential requirement. To this end, it is desirable for a robot to understand the environment at both geometric and semantic levels. Therefore in recent years, research towards place classification has been gaining in popularity. After the era of heuristic and rulebased approaches, supervised learning algorithms have been extensively used for this purpose, showing satisfactory performance levels. However, most of those approaches have only been trained and tested in the same environments and thus impede a generalized solution. In this paper, we have proposed a semisupervised place classification over a generalized Voronoi graph (SPCoGVG) which is a semi-supervised learning framework comprised of three techniques: support vector machine (SVM), conditional random field (CRF) and generalized Voronoi graph (GVG), in order to improve the generalizability. The inherent problem of training CRF with partially labeled data has been solved using a novel parameter estimation algorithm. The effectiveness of the proposed algorithm is validated through extensive analysis of data collected in international university environments.
Nguyen, L, Miro, JV, Shi, L & Vidal-Calleja, T 2019, 'Gaussian Mixture Marginal Distributions for Modelling Remaining Pipe Wall Thickness of Critical Water Mains in Non-Destructive Evaluation', 2019 IEEE International Conference on Cybernetics and Intelligent Systems (CIS) and IEEE Conference on Robotics, Automation and Mechatronics (RAM), IEEE International Conference on Cybernetics and Intelligent Systems, and Robotics, Automation and Mechatronics, IEEE, Bangkok, Thailand.View/Download from: Publisher's site
Rapidly estimating the remaining wall thickness (RWT) is paramount for the non-destructive condition assessment evaluation of large critical metallic pipelines. A robotic vehicle with embedded magnetism-based sensors has been developed to traverse the inside of a pipeline and conduct inspections at the location of a break. However its sensing speed is constrained by the magnetic principle of operation, thus slowing down the overall operation in seeking dense RWT mapping. To ameliorate this drawback, this work proposes the partial scanning of the pipe and then employing Gaussian Processes (GPs) to infer RWT at the unseen pipe sections. Since GP prediction assumes to have normally distributed input data - which does correspond with real RWT measurements - Gaussian mixture (GM) models are proven in this work as fitting marginal distributions to effectively capture the probability of any RWT value in the inspected data. The effectiveness of the proposed approach is extensively validated from real-world data collected in collaboration with a water utility from a cast iron water main pipeline in Sydney, Australia.
Shi, L, Valls Miro, J, Vidal Calleja, T, Vitanage, D & Rajalingam, J 2017, 'Innovative Data-driven "along-the-pipe" Condition Assessment for Critical Water Mains', OZWATER'17 Australia's International Water Conference & Exhibition, OZWATER'17 Australia's International Water Conference & Exhibition, Australian Water Association, Sydney, pp. 1-8.
Recent research findings on remaining life prediction for older Cast Iron critical water mains suggest increasing reliability by calculating stress concentration factors from the corrosion patch geometries expected to be present in the asset, not just extreme pitting as is generally carried out within the industry. This study proposes an innovative data-driven "along-the-pipe" framework able to utilise local inspection results further by capturing data correlations present in the remaining wall thickness measurement. This knowledge can in turn be utilised to produce estimates for "along-the-pipe" patch geometry predictions, hence remaining life. Results from inspections in a real pipeline in the Sydney Water network are compared to conventional Extreme Value Analysis (EVA) to validate the improvements of the proposed strategy.
Song, B, Wang, H, Xiao, W, Shi, L & Huang, S 2017, 'Gaussian Process Model Enabled Particle Filter for Device-Free Localization', 2017 20th International Conference on Information Fusion (Fusion), International Conference on Information Fusion, IEEE, Xi'an, China, pp. 1-7.View/Download from: Publisher's site
Device-free localization (DFL) is an emerging wireless network target localization technique that does not need to attach any electronic device with the target. It is remaining as a challenging research problem due to the weak wireless signals and the uncertain wireless communication environment. In this paper, a novel Gaussian Process (GP) based wireless propagation model is proposed to describe the likelihood relationship between the target location and the changes of the RSS measurement for a wireless link. Sequentially Particle Filter (PF) is applied to the DFL for estimating the location of the target, after the GP model is trained using the experimental measurements of the link. Experimental results demonstrate that the proposed GP-PF algorithm can track the target with much better localization accuracy than the Support Vector Machine (SVM) based PF approach.
Wang, M, Su, D, Shi, L, Liu, Y & Miro, JV 2017, 'Real-time 3D human tracking for mobile robots with multisensors', Proceedings - IEEE International Conference on Robotics and Automation, pp. 5081-5087.View/Download from: Publisher's site
© 2017 IEEE. Acquiring the accurate 3-D position of a target person around a robot provides fundamental and valuable information that is applicable to a wide range of robotic tasks, including home service, navigation and entertainment. This paper presents a real-time robotic 3-D human tracking system which combines a monocular camera with an ultrasonic sensor by the extended Kalman filter (EKF). The proposed system consists of three sub-modules: monocular camera sensor tracking model, ultrasonic sensor tracking model and multi-sensor fusion. An improved visual tracking algorithm is presented to provide partial location estimation (2-D). The algorithm is designed to overcome severe occlusions, scale variation, target missing and achieve robust re-detection. The scale accuracy is further enhanced by the estimated 3-D information. An ultrasonic sensor array is employed to provide the range information from the target person to the robot and Gaussian Process Regression is used for partial location estimation (2-D). EKF is adopted to sequentially process multiple, heterogeneous measurements arriving in an asynchronous order from the vision sensor and the ultrasonic sensor separately. In the experiments, the proposed tracking system is tested in both simulation platform and actual mobile robot for various indoor and outdoor scenes. The experimental results show the superior performance of the 3-D tracking system in terms of both the accuracy and robustness.
Wang, M, Su, D, Shi, L, Liu, Y & Valls Miro, J 2017, 'Real-Time 3D Human Tracking for Mobile Robots with Multisensors', ICRA 2017 Program, IEEE International Conference on Robotics and Automation, IEEE, Singapore, Singapore.View/Download from: Publisher's site
Acquiring the accurate 3-D position of a target person around a robot provides fundamental and valuable information that is applicable to a wide range of robotic tasks, including home service, navigation and entertainment. This paper presents a real-time robotic 3-D human tracking system which combines a monocular camera with an ultrasonic sensor by the extended Kalman filter (EKF). The proposed system consists of three sub-modules: monocular camera sensor tracking model, ultrasonic sensor tracking model and multi-sensor fusion. An improved visual tracking algorithm is presented to provide partial location estimation (2-D). The algorithm is designed to overcome severe occlusions, scale variation, target missing and achieve robust re-detection. The scale accuracy is further enhanced by the estimated 3-D information. An ultrasonic sensor array is employed to provide the range information from the target person to the robot and Gaussian Process Regression is used for partial location estimation (2-D). EKF is adopted to sequentially process multiple, heterogeneous measurements arriving in an asynchronous order from the vision sensor and the ultrasonic sensor separately. In the experiments, the proposed tracking system is tested in both simulation platform and actual mobile robot for various indoor and outdoor scenes. The experimental results show the superior performance of the 3-D tracking system in terms of both the accuracy and robustness.
Liao, Y, Kodagoda, S, Wang, Y, Shi, L & Liu, Y 2016, 'Understand scene categories by objects: A semantic regularized scene classifier using Convolutional Neural Networks', 2016 IEEE International Conference on Robotics and Automation (ICRA), IEEE International Conference on Robotics and Automation, IEEE, Stockholm, Sweden, pp. 2318-2325.View/Download from: Publisher's site
Scene classification is a fundamental perception task for environmental understanding in today's robotics. In this paper, we have attempted to exploit the use of popular machine learning technique of deep learning to enhance scene understanding, particularly in robotics applications. As scene images have larger diversity than the iconic object images, it is more challenging for deep learning methods to automatically learn features from scene images with less samples. Inspired by human scene understanding based on object knowledge, we address the problem of scene classification by encouraging deep neural networks to incorporate object-level information. This is implemented with a regularization of semantic segmentation. With only 5 thousand training images, as opposed to 2.5 million images, we show the proposed deep architecture achieves superior scene classification results to the state-of-the-art on a publicly available SUN RGB-D dataset. In addition, performance of semantic segmentation, the regularizer, also reaches a new record with refinement derived from predicted scene labels. Finally, we apply our model trained on SUN RGB-D dataset to a set of images captured in our university using a mobile robot, demonstrating the generalization ability of the proposed algorithm.
Ma, H, Shi, L, Kodagoda, S & Xiong, R 2016, 'A Semantic Labeling Strategy to Reject Unknown Objects in Large Scale 3D Point Clouds', Proceedings of the 35th Chinese Control Conference, Chinese Control Conference, IEEE, Chengdu, Sichuan, China, pp. 7070-7075.View/Download from: Publisher's site
In recent years, there has been a growing interest in the research of semantic labeling of indoor scenes represented by 3D point clouds. A fundamental problem that has largely been oversighted in the current research is the way of dealing with the unknown class which collectively includes all the objects that are of no interest to the application developer. In the training stage, these objects are either completely removed or labeled as unknown, resulting in a trained model which is not fully and fairly exposed to the actual sample space. In the test stage, the unknown objects are naturally present and provided to the classifier, causing a significant drop of the classification accuracy-usually 20%~30%. Simply improving the features or the classifier will not address the root cause problem. In this paper, we propose a labeling framework combining both Conditional Random Field (CRF) and PI-SVM to specifically solve the problem caused by the unknown class. First, we use a CRF to model the contextual relations in the 3D space, for which the parameters for both node potential and edge potential are learned from training data. Then, we make use of the rejection strategy of the PI-SVM, which estimates an unnormalized probability for each class. Finally, we reinforce the result of CRF with the belief provided by the PI-SVM, and the labeling result is based on the agreement of the two classifiers. The proposed method takes advantage of the global optimization of CRF and the advantage of unknown rejection of PI-SVM. Experimental results on publicly available data set show that this method has improved the classification accuracy by 10.7% given the accuracy drop of 19.23% caused by the unknown.
Shi, L & Valls Miro, J 2016, 'Towards Optimized and Reconstructable Sampling Inspection of Pipe Integrity for Improved Efficiency of NDT', 2016 IWA World Water Congress, Brisbane, Queensland, Australia.
Shi, L, Valls Miro, J, Rajalingam, J, Wood, R & Vitanage, D 2016, 'High Precision GPS Aided In-pipe Distance Calibration For Satellite Image-based Pipeline Mapping', OZWATER'16 Australia's International Water Conference & Exhibition, OZWATER'16 Australia's International Water Conference & Exhibition, Australian Water Association, Melbourne, pp. 1-8.
Asset management and pipe condition assessment (CA) activities in the water industry usually require locating buried pipes accurately to minimise inspection and maintenance costs. A typical
challenge in practice is locating an anomaly detected by an in-pipe inspection tool from aboveground in order to dig up a pipe for replacement. Accumulated in-pipe errors over longer distances in particular can easily lead to selecting the wrong pipe section for further investigation or exhumation. In fact, some in-pipe CA providers suggest utility personnel dig up a number of sections of pipe around the suggested location so as to ensure finding the target section. In this paper we propose a mechanism to accurately correlate a 3D pipeline profile built from GPS surveying results of aboveground pipeline features with in-pipe chainage distances, so as to establish an accurate link between above-ground GPS coordinates and inpipe distance measurements. This approach naturally characterises and corrects for some of the most prominent in-pipe chainage measurement errors that can lead to uncertainties about the reported location of a buried pipeline from above-ground. The detailed pipeline information can then be projected onto satellite imagery as an accurate easy-to-understand reference for efficient decision making.
Shi, L, Valls Miro, J, Zhang, T, Vidal Calleja, T, Sun, L & Dissanayake, G 2016, 'Constrained Sampling of 2.5D Probabilistic Maps for Augmented Inference', IEEE International Conference on Intelligent Robots and Systems, IEEE/RSJ International Conference on Intelligent Robots and Systems, IEEE, Daejeon, Korea.View/Download from: Publisher's site
Zhang, T, Huang, S, Liu, D, Shi, L, Zhou, C & Xiong, R 2016, 'A Method of State Estimation for Underwater Vehicle Navigation Around A Cylindrical Structure', 2016 IEEE 11th Conference on Industrial Electronics and Applications (ICIEA), IEEE Conference on Industrial Electronics and Applications, IEEE, Hefei, China, pp. 101-106.View/Download from: Publisher's site
Recently, increasing efforts have been focused on the development and adoption of autonomous underwater vehicles (AUVs) for various applications. However, the GPS signals are usually unavailable, the vehicle dynamics is very uncertain, and the complicated vision based localization algorithms may not work well in the underwater environments. Hence, accurate and timely state estimation using low-cost sensors remains a challenge for the control and navigation of AUVs. This paper considers the state estimation problem for underwater vehicle navigation around a cylindrical structure. The vehicle is assumed to be equipped with only low-cost sensors: an inertia measurement unit (IMU), a pressure sensor and a monocular camera. By exploiting the prior knowledge on the size and shape of the structure, an efficient algorithm for estimating the state of the AUV is developed without using any dynamic model. Firstly, a state observer is proposed under the condition that the localization result (rotational and translational position) is available. Next, we present a method for localization based on the IMU readings, pressure sensor readings and the image of the cylindrical structure, which uses the geometry of the structure and only requires simple image processing (line extraction). Then we prove that the proposed observer is globally stable. Preliminary experimental results and simulation results are reasonable and promising, which implies the proposed method has potential to be used in the real AUV navigation applications.
Alvarez, JK, Abeywardena, DA, Shi, LS & Kodagoda, SK 2015, 'Using Hidden Markov Models to Improve Floor Level Localization', Website Proceedings of the Australasian Conference on Robotics and Automation 2015, Australasian Conference on Robotics and Automation, ARAA, Canberra.
Shi, L, Sun, L, Vidal Calleja, T & Miro, JV 2015, 'Kernel-Specific Gaussian Process for Predicting Pipe Wall Thickness Maps', Website Proceedings of the Australasian Conference on Robotics and Automation 2015, Australasian Conference on Robotics and Automation, AARA, Canberra, pp. 1-8.
Data organised in 2.5D such as elevation and
thickness maps has been extensively studied in
the fields of robotics and geostatistics. These
maps are typically a probabilistic 2D grid that
stores an estimated value (height or thickness)
for each cell. Modelling the spatial dependencies
and making inference on new grid locations
is a common task that has been addressed using
Gaussian random fields. However, inference
faraway from the training areas results quite
uncertain, therefore not informative enough for
some applications. The objective of this research
is to model the status of a pipeline based
on limited and sparse local assessments, predicting
the likely condition on pipes that have
not been inspected. A customised kernel for
Gaussian Processes (GP) is proposed to capture
the spatial correlation of the pipe wall
thickness data. An estimate of the likely condition
of non-inspected pipes is achieved by concretising
GP to a multivariate Gaussian distribution
and generating realisations from the distribution.
The performance of this approach is
evaluated on various thickness maps from the
same pipeline, where data have been obtained
by measuring the actual remaining wall thickness.
The output of this work aims to serve as
the input of
Shi, L, Kodagoda, S & Piccardi, M 2013, 'Towards Simultaneous Place Classification and Object Detection based on Conditional Random Field with Multiple Cues', 2013 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), IEEE/RSJ International Conference on Intelligent Robots and Systems, IEEE, Tokyo, Japan, pp. 2806-2811.View/Download from: Publisher's site
Khushaba, RN, Shi, L & Kodagoda, S 2012, 'Time-Dependent Spectral Features for Limb Position Invariant Myoelectric Pattern Recognition', The 12th International Symposium on Communications and Information Technologies (ISCIT), International Symposium on Communications and Information Technologies, IEEE, Surfers Paradise, Gold Coast, Australia, pp. 1020-1025.View/Download from: Publisher's site
Recent studies on the myoelectric control of powered prosthetics revealed several factors that affect its clinical performance. One of the important factors is the variation in the limb position associated with normal use which can have a substantial impact on the robustness of Electromyogram (EMG) pattern recognition. To solve this problem, we propose in this paper a new feature extraction algorithm based on set of spectral moments that extracts the relevant information about the EMG power spectrum in an accurate and efficient manner. The main goal is to rely on effective knowledge discovery and pattern recognition methods to discover the neural information embedded in the EMG signals regardless of the limb position. Specifically, the proposed features define descriptive qualities for the general time domain-based characterization of the EMG spectral amplitude, spectral sparsity, and irregularity factor by the application of mathematical-statistical methods which also include frequency consideration. The performance of the proposed spectral moments is tested on EMG data collected from eight subjects, while implementing eight classes of movements, each at five different limb positions. Practical results indicate that training the classifier on the EMG moments collected from multiple positions and testing on completely unseen positions can achieve significant reduction in the classification error rates of upon â10% on average across all subjects and limb ositions.
Shi, L, Kodagoda, S & Dissanayake, G 2012, 'Application of Semi-supervised Learning with Voronoi Graph for Place Classification', 2012 IEEE/RSJ International Conference on Intelligent Robots and Systems, IEEE/RSJ International Conference on Intelligent Robots and Systems, IEEE, Vilamoura, Algarve, Portugal, pp. 2991-2996.View/Download from: Publisher's site
Representation of spaces including both geometric and semantic information enables a robot to perform high-level tasks in complex environments. Therefore, in recent years identifying and semantically labeling the environments based on onboard sensors has become an important competency for mobile robots. Supervised learning algorithms have been extensively used for this purpose with SVM-based solutions showing good generalization properties. The CRF-based approaches take the advantage of connectivity information of samples thereby provide a mechanism to capture complex dependencies. Blending the complementary strengths of Support Vector Machine (SVM) and Conditional Random Field (CRF), there have been algorithms to exploit the advantages of both to enhance the overall accuracy of place classification in indoor environments. However, experiments show that none of the above approaches deal well with diversified testing data. In this paper, we focus mainly on the generalization ability of the model and propose a semi-supervised learning strategy, which essentially improves the performance of the system. Experiments have been carried out on six real-world maps from different universities around the world and the results from rigorous testing demonstrate the feasibility of the approach.
Shi, L, Kodagoda, S, Khushaba, RN & Dissanayake, G 2012, 'Application of CRF and SVM based Semi-supervised Learning for Semantic Labeling of Environments', 2012 12th International Conference on Control, Automation, Robotics & Vision, International Conference on Control, Automation, Robotics and Vision, IEEE, Guangzhou, pp. 835-840.View/Download from: Publisher's site
Understanding the environment in both geometric and semantic levels enables a robot to perform high-level tasks in complex environments. Therefore in recent years research towards identifying and semantically labeling the environments based on onboard sensors for mobile robots has been gaining popularity. After the era of heuristic and rule-based approaches, supervised learning algorithms like Support Vector Machines (SVM) and AdaBoost have been extensively used for this purpose showing satisfactory performance. With the introduction of graphical models, approaches like Conditional Random Fields (CRF) which take the advantage of connectivity of samples provide more flexibility to capture complex dependencies. In this paper, we focus on a real-world task which challenges the generalization ability of the model, evaluate some graph based features, propose a semi-supervised learning algorithm by iteratively utilizing the results from SVM and CRF, and suggest a solution for CRF parameter estimation with partially labeled training data. Experiments have been conducted on six realworld indoor environments demonstrating the competence of the algorithm.
Shi, L, Kodagoda, S & Ranasinghe, R 2011, 'Fast Indoor Scene Classification Using 3D Point Clouds', Australasian Conference on Robotics and Automation 2011, Australasian Conference on Robotics and Automation, The ACRA 2011 Organising Committee, Monash University, Melbourne Australia, pp. 1-7.
A representation of space that includes both geometric and semantic information enables a robot to perform high-level tasks in complex environments. Identifying and categorizing environments based on onboard sensors are essential in these scenarios. The Kinectâ¢, a 3D low cost sensor is appealing in these scenarios as it can provide rich information. The downside is the presence of large amount of information, which could lead to higher computational complexity. In this paper, we propose a methodology to efficiently classify indoor environments into semantic categories using Kinectâ¢ data. With a fast feature extraction method along with an efficient feature selection algorithm (DEFS) and, support vector machines (SVM) classifier, we could realize a fast scene classification algorithm. Experimental results in an indoor scenario are presented including comparisons with its counterpart of commonly available 2D laser range finder data.
Shi, L, Kodagoda, S & Dissanayake, G 2010, 'Environment Classification and Semantic Grid Map Building Based on Laser Range Finder Data', IROS 2010 Workshop on Semantic Mapping and Autonomous Knowledge Acquisition, Workshop on Semantic Mapping and Autonomous Knowledge Acquisition, online proceeding of IROS 2010 workshop, Taipei, pp. 1-6.
Human robot interaction has become an important area of research in the robotics community. High level abstractions, which are commonly used by humans, can be learnt by robots to effectively communicate with humans. In this paper, we propose a Semantic Grid Map (SGM) to represent an environment. SGM is similar to an Occupancy Grid (OG) map, however with high level information as environment type labels. We use a robot-mounted laser range finder (LRF) data to learn and classify an environment into various area types. Then the classification results are combined probabilistically to update the semantic grid map. The classification accuracy is further improved by outlier rejection and topological correction. Finally we present a labeling strategy while a robot is exploring an unknown environment. Experimental results of a robot exploring in a university environment are presented to assess the performance of the algorithm.
Shi, L, Kodagoda, S & Dissanayake, G 2010, 'Laser Range Data Based Semantic Labeling of Places', Proceedings of the IEEE/RSJ International Conference on Intelligent Robots and Systems, IEEE/RSJ International Conference on Intelligent Robots and Systems, IEEE, Taipei, Taiwan, pp. 5941-5946.View/Download from: Publisher's site
Extending metric space representations of an environment with other high level information, such as semantic and topological representations enable a robotic device to efficiently operate in complex environments. This paper proposes a methodology for a robot to classify indoor environments into semantic categories. Classification task, using data collected from a laser range finder, is achieved by a machine learning approach based on the logistic regression algorithm. The classification is followed by a probabilistic temporal update of the semantic labels of places. The innovation here is that the new algorithm is able to classify parts of a single laser scan into different semantic labels rather than the conventional approach of gross categorization of locations based on the whole laser scan. We demonstrate the effectiveness of the algorithm using a data set available in the public domain.
Shi, L, Kodagoda, S & Dissanayake, G 2010, 'Multi-class Classification for Semantic Labeling of Places', Proceedings of the 11th International Conference on Control, Automation, Robotics and Vision (ICARCV 2010), International Conference on Control, Automation, Robotics & Vision, IEEE, Singapore, pp. 2307-2312.View/Download from: Publisher's site
AbstractâHuman robot interaction is an emerging area of research, where human understandable robotic representations can play a major role. Knowledge of semantic labels of places can be used to effectively communicate with people and to develop efficient navigation solutions in complex environments. In this paper, we propose a new approach that enables a robot to learn and classify observations in an indoor environment using a labeled semantic grid map, which is similar to an Occupancy Grid like representation. Classification of the places based on data collected by laser range finder (LRF) is achieved through a machine learning approach, which implements logistic regression as a multi-class classifier. The classifier output is probabilistically fused using independent opinion pool strategy. Appealing experimental results are presented based on a data set gathered in various indoor scenarios.
Shi, L, Kodagoda, S & Dissanayake, G 2010, 'Semantic Grid Map Building', Proceedings of the Australasian Conference on Robotics and Automation 2010 (ACRA 2010), Proceedings of the Australasian Conference on Robotics and Automation, Australasian Conference on Robotics and Automation, Brisbane, Australia, pp. 1-7.
Conventional Occupancy Grid (OG) map which contains occupied and unoccupied cells can be enhanced by incorporating semantic labels of places to build semantic grid map. Map with semantic information is more understandable to humans and hence can be used for efficient communication, leading to effective human robot interactions. This paper proposes a new approach that enables a robot to explore an indoor environment to build an occupancy grid map and then perform semantic labeling to generate a semantic grid map. Geometrical information is obtained by classifying the places into three different semantic classes based on data collected by a 2D laser range finder. Classification is achieved by implementing logistic regression as a multi-class classifier, and the results are combined in a probabilistic framework. Labeling accuracy is further improved by topological correction on robot position map which is an intermediate product, and also by outlier removal process on semantic grid map. Simulation on data collected in a university environment shows appealing results.