Artificial Intelligence and Robotics

The 3D computer vision application become popular in recent years and estimating the object distance is basic technology. This study used laser array to beam the object then generate highlight characteristic point, and then applied Fuzzy C-mean (FCM) and Ant colony (ACO) to classify characteristic points on image. Finally, used conjugate pair and characteristic point on object and then based on Epipolar plane the object distance was estimated those maximum error rate is $$ \pm 5.6{\% } $$.

The acupuncture points are given acupuncture or acupressure so to stimulate the meridians on each corresponding internal organs with a treatment of physical illness. The goal of this study is to use image technique to automatically find acupuncture positions of a palm to help non-related professionals can clearly identify the location of acupuncture points on him palm. In this paper, we use the skin color detection, color transform, edge detection, histogram and fast packet method to extract the palm and find out the acupunctures. First, we use fast packet method to get the acupunctures of the finger. And then a histogram technique was used to obtain the acupuncture points of the valley of the fingers. Finally, the valley points and fingertips of the finger are used as a reference combined with the standard deviation of data images to calculate the position of the palm acupuncture points. From the simulation result, it is demonstrated that our design is an effective method for indicating the acupuncture points of a palm.

In this paper, we propose a low-rank representation and locality-constrained regression (LLRLCR) based approach to learn the occlusion-robust discriminative representations features for low-resolution face recognition tasks. For gallery set, LLRLCR uses double low-rank representation to reveal the underlying data structures; for probe set, LLRLCR uses locality-constrained matrix regression to learn discriminative representation features robustly. The proposed method allows us to fully exploit the structure information in gallery and probe data simultaneously. Finally, after getting the resolution-robust features, a simple yet powerful sparse representation based classifier engine is used to predict the face labels. Experiments conducted on the AR database with occlusions have shown that the proposed method can obtain promising recognition performance than many state-of-the-art LR face recognition approaches.

With the development of deep neural networks, researchers have developed lots of algorithms related to face and achieved comparable results to human-level performance on several databases. However, few feature extraction models work well in the real world when the subject which is to be recognized has limited samples, for example, only one ID photo can be obtained before the face recognition task. To our best knowledge, there is no face database which contains ID photos and pictures from the real world for a subject simultaneously. To fill this gap, we collected 100 celebrities’ ID photos and their about 1000 stills or life pictures and formed a face database called FDID. Besides, we proposed a novel face recognition algorithm and evaluated it with this new database on the real-life videos.

Image relighting is to change the illumination of an image to a target illumination effect without known the original scene geometry, material information and illumination condition. We propose a novel outdoor scene relighting method, which needs only a single reference image and is based on material constrained layer decomposition. Firstly, the material map is extracted from the input image. Then, the reference image is warped to the input image through patch match based image warping. Lastly, the input image is relighted using material constrained layer decomposition. The experimental results reveal that our method can produce similar illumination effect as that of the reference image on the input image using only a single reference image.

An adaptive reinforcement learning method designed to facilitate the on-line lifting motion of the human forearm is here proposed. Its purpose is to use the control based on the proposed learning method to perform the lifting motion. The learning algorithm is an actor-critic learning based on the neural network that used the normalized radial basis function. The paper shows a simulation of the motion of the forearm lifting process. As shown in the results, the forearm continues to lift from a horizontal position to a vertical position. During this process, both the state space and action space are continuous.

In the Iaas service mode for cloud computing, cloud providers allocate resources in the form of Virtual Machines (VM) to cloud users via auction mechanism. The existing auction mechanism lacks self-adapting adjustment to market changes. An improved online auction mechanism by taking into account the changes in demand during peak and trough period in the allocation scheme has been proposed, so that the auctioneer can make decisions reasonably, improve resource utilization rate, and bring higher profits. Firstly, we present an auction framework for VM allocation based on multi-time period, then prove the mechanism satisfies individual rationality and incentive compatibility. Finally, we try to use the real workload file to perform simulation experiments to verify the effectiveness of the improved online mechanism.

In this paper, we proposed a joint hierarchy model to represent the motion of human according to the covariance feature of adjacent joints using Kinect. SVM is used for the action classification. Experimental results show that the proposed model improves the recognition accuracy with less computation complexity.

Medical path varies due to the changes in external factors, and the key to the changes is the decision-making in symptomatic treatment by medical experts; how to make full use of historical medical data and recommend high quality treatment options to new cases under the current circumstance of data explosion, is the research focus of this paper. This paper first presents the standard for medical path based on cloud platform, going into the definition of model of medical services; finally, the medical optimization factor is given to realize the evolution of online medical program of a disease based on QoS.

In order to reduce the noise points when measure the shape of 3D object using Phase Shift Profilometry (PSP) methods, in this paper we propose a novel Three Wave Length PSP (TWPSP) method. Firstly, the direct problems of equivalent wavelength and unwrapped phase are analyzed. Then, the solution of unwrapping method for TWPSP is derived. Finally, based on the global phase filtering the phase noises are reduced. We simulated and compared the proposed TWPSP and the classical TWPSP. Experiential results shown that the noises are greatly restrained. Since the proposed method does not need to calculate the equivalent phase maps, the implement is improved than the classical TWPSP.

Magnetic detection electrical impedance tomography (MD-EIT) can be used to reconstruct images of conductivity from magnetic field measurement taken around the body in vivo. To achieve the lung tumor imaging based on the MD-EIT, this study established the human thoracic model according to CT image. The Moore-Penrose inverse, TSVD and Tikhonov regularization were tried before the reconstruction to improve imaging accuracy respectively. Four levels of noise were added into the simulation data to meet the real detection situation (the SNR is 30, 60, 90 and 120 dB). The reconstruction results, which were pre-processed by TSVD regularization, had the best performance. The average relative error (ARE) values of current density distribution equals 0.21 and 0.22 for healthy and lung tumor person respectively. The tumor in lung can be distinguished clearly from the MD-EIT image. The MD-EIT is one of the most promising technology in dynamic lung imaging for clinical application.

Road detection aims at detecting the (drivable) road surface ahead vehicle and plays a crucial role in driver assistance system. To improve the accuracy and robustness of road detection approaches in complex environments, a new road detection method based on CNN (convolutional neural network) and MRF (markov random field) is proposed. The original road image is segmented into super-pixels of uniform size using simple linear iterative clustering (SLIC) algorithm. On this basis, we train the CNN which can automatically learn the features that are most beneficial to classification. Then, the trained CNN is applied to classify road region and non-road region. Finally, based on the relationship between the super-pixels neighborhood, we utilize MRF to optimize the classification results of CNN. Quantitative and qualitative experiments on the publicly datasets demonstrate that the proposed method is robust in complex environments. Furthermore, compared with state-of-the-art algorithms, the approach provides the better performance.

To increase the classification accuracy and stability of the prediction model, an approach to evaluate the quality of samples’ hyperspectral image is needed. The spectral correlation analysis of each pixel was used to determine quality of the sample’s hyperspectral image in this study. 400 hyperspectral image ROIs were extracted from 20 apples (10 apples with waxed and the other 10 apples without any waxed) and the data were separated into 300 as train set and 100 as test set randomly. The experimental group data were evaluated by the spectral correlation analysis, and only qualified data were used for model training. The control group data were all used for modeling training. The least squares support vector machine (LS-SVM) model were used to establish the classification model between the hyperspectral image and waxed situation. The prediction result showed the classification accuracy were 94% and 86% when the low-quality sample data for training were filtered by spectral correlation analysis. By evaluating the quality of the hyperspectral image measured, more reliable prediction results can be obtained, which can make the noninvasive discrimination of food safety come to the practice application sooner.

Automatic segmentation of organs from medical images is indispensable for the applications of computer-aided diagnosis (CAD) and computer-assisted surgery (CAS). Statistical Shape Models (SSMs) based scheme have been proved as the accurate and robust methods for extraction of anatomical structures. A key step of this approach is the need to place the sampled points(landmarks) with correspondence across the training set. On the one hand, the correspondence of landmarks is related the quality of SSMs. On the other hand, in many cases the location of key landmarks should be manipulated by physicians, since an unattended system is hard to use in most clinical applications. In this paper, we establish a dense correspondence across the whole training set automatically by surface features, which are registered using diffeomorphic demons approach. And the optimization is executed on spherical domain. We establish the SSM for lung regions, the deformation of where is greatly. Finally, we derive quantitative measures of model quality and comparison of segmentation results using the model with non-optimized correspondence.

To control humanoid robot, one of the challenges is to keep the balance. In this paper, a joint angle-based control (JAC) scheme is proposed for the lower-body of humanoid robot NAO to imitate human motion via Kinect. To keep the balance, the joint angles in the lower-body of NAO are optimized according to the tracked human motion and the current state of the robot. Experimental results show that the control scheme works efficiently even when the humanoid robot performs the complex movement such as standing on single foot.

A shoe-mounted indoor positioning system is designed based on the low-cost MEMS (Micro-Electro-Mechanical System) inertial measurement unit (IMU). To solve the problem of the high noise and drift of the MEMS inertial sensor, the method of the coarse calibration for MIMU is studied. Aiming at solving the shortcomings of the traditional zero velocity interval detection algorithm with single threshold, an alternative algorithm for the zero velocity interval detection based on multiple gaits is designed, which judges the motion state before detection. Then, the Kalman filter algorithm based on the zero velocity update (ZUPT) and zero angular rate update (ZARU) is designed to estimate and compensate the cumulative error of the sensors during walking. Finally, a field test based on a low-cost MEMS IMU is carried out with four different gaits, slow walking, up and down staircase, striding forward and long-distance walking with variable speed. The results show that the positioning error of the proposed method is only 3% of the walking distance.

With the development of economic society, comprehensive fitness has risen to national strategy, and people’s demand for physical training is getting bigger and bigger. In the information age, traditional stadiums can not meet the needs of people for sports, so the intelligent stadium is the inevitable trend of future development. Also, the positioning of the human in the stadium is the basis for the realization of human-computer interaction and system function. This paper first analyzes the positioning requirements in the intelligent stadium, then puts forward the necessity of the positioning system in the intelligent stadium. The WIFI location fingerprint localization algorithm is further studied and improved while the application of the algorithm in the positioning system of the intelligent stadium is explained. Finally, we completed data collection and real-time positioning in the stadium, designed and implemented the LBS location service based on hospital.

Obstacles such as ramps, steps and irregular floor surfaces are commonly encountered in homes, offices and other public spaces. These obstacles frequently limit the daily activities of people who use mobility aids. To expand the scope of their activities, a wheel mechanism for climbing a step while reducing the horizontal climbing force is proposed to be applied into self-propelled wheelchairs. The physical and mental burdens of caregivers and medical staff can be reduced by making the users of the mechanism more self-sufficient. Specifically, the proposed step-climbing mechanism for the self-propelled wheelchair relies on offsetting the rotational axis of the wheel for step climbing. This paper provides details on this offsetting mechanism and its force analyses.

How to make good use of different kinds of security resources to protect the urban city has always been an important social problem. Firstly, we propose a multi-agent framework in which various kinds of agents cooperate to complete tasks. In this framework, we develop an auction-based task allocation mechanism for police force agents. Secondly we come up with a double oracle algorithm for policing resources allocation in cooperative security games on graphs. Lastly, according to the experimental results, our algorithm runs approximately 25% faster than the former work.

Global calibration has direct influence on measurement accuracy of multi-camera system. The present calibration methods are hard to be applied in field calibration for the usage of complicated structures with accurate geometric or simple parts requiring overlapping view field of the system. Aiming at these problems, a new method is proposed in this paper by using two fixed plane targets with invariable pose. Objective functions are established according to constantness of the distance between original points and the axis angles of the plane targets, and nonlinear optimization is improved by means of Rodrigues transform. An apparatus is manufactured for real calibration experiments, and results verify the effectively and reliability of the method.

In this paper, a regression analysis method based on the combination of Least Angle Regression (LAR) and Partial Least Squares (PLS) is proposed, which uses the non-invasive characteristics of near infrared spectroscopy (NIRS) to implement early screening of leukemia patients. First, the LAR method is used to eliminate collinearity between variables, second, PLS is employed to further build model for the wavelengths which are selected by the LAR. The result shows that this method needs less wavelength points and has more excellent performance in correlation coefficient and root mean square error, that are 0.9492 and 0.5917 respectively. The comparison experiments demonstrate that the LAR-PLS regression model has an advantage over principal component regression (PCR), the LAR-PCR regression model, successive projections algorithm (SPA) and elimination of uninformative variables (UVE) combined with PLS method in terms of predictive accuracy for screening leukemia patients.

This paper describes a method for drug discrimination with near infrared spectroscopy based on SSAE-ELM. ELM instead of the BP was introduced to fine-tuning SSAE, which can reduce the training time of SSAE and improve the practical application of the deep learning network. The work in the paper used near infrared diffuse reflectance spectroscopy to identify Aluminum-plastic packaging of cefixime tablets drugs from different manufacturers as examples to verify the proposed method. Specifically, we adopted SSAE-ELM to binary and multi-class classification discriminations with different sizes of drug dataset. Extensive experiments were conducted to compare the performances of the proposed method with ELM, BP, SVM and SWELM. The results indicate that the proposed method not only can obtain high discrimination accuracy with superior stability but also reduce the training time of SSAE in binary and multi-class classification. Therefore, the SSAE-ELM classifier can achieve an optimal and generalized solution for spectroscopy identification.

With the explosive growth of information on the Web, Semantic Web and related technologies such as linked data and commonsense knowledge bases, have been introduced. ConceptNet is a commonsense knowledge base, which is available for public use in CSV and JSON format; it provides a semantic graph that describes general human knowledge and how it is expressed in natural language. Recently, an RDF presentation of ConceptNet called ConceptRDF has been proposed for better use in different fields; however, it has some problems (e.g., information of concepts is sometimes misexpressed) caused by the improper conversion model. In this paper, we propose a concise conversion model to improve the RDF expression of ConceptNet. We convert the ConceptNet into RDF format and perform some experiments with the conversion results. The experimental results show that our conversion model can fully express the information of ConceptNet, which is suitable for developing many intelligent applications.

In view of the consumption and distribution of resources in the world, the necessity of deep-sea mining video image processing is illustrated. Based on a large number of relevant literatures, this paper summarizes the research status of image processing methods for deep-sea mining observation video system, introduces the advantages of the improved median filter algorithm, the improved dark channel prior algorithm and the improved nonlocal mean denoising method. Some problems of the original methods are analyzed. These aim to provide reference for the optimization and improvement of image processing methods for deep-sea mining video observation.

The key concept of IoT has the potential to be both a benefit and a threat not only to the ICT industry but also to the consumer electronics and device industries. Panasonic has high expectations for the IoT business and has already started providing value to individuals who live in smart homes. From a business viewpoint, we can simply describe “IoT” as ICT technologies that expanded from the computer to the non-computer world. However, there remain various technological issues to realize a genuine IoT world. This lecture introduces IoT business activities in the consumer electronics industry, and is followed by a discussion of the technological challenges for IoT both in the software and hardware layers.

Recently years, convolutional neural networks (CNN) has been a hot spot in various areas such as object detection, classification. As deep study in CNN, its performance is almost human-competitive. We find that the test accuracy largely depends on the relationship of samples in feature space. Softmax loss is widely used in many deep learning algorithms. However, it cannot directly reflect this kind of relationship. In this paper, we design a new loss function, named inter loss. This inter loss function can maximizes the distance between different classes, analogous to maximizing margin in SVM. By integrating inter loss and softmax loss, larger inter-class distance and smaller intra-class distance can be obtained. In this way, we can significantly improve the accuracy in classification. Impressive results is obtained in SVHN and CIFAR-10 datasets. However, our main goal is to introduce a novel loss function tasks rather than beating the state-of-the-art. In our experiments, other forms of loss functions based on inter and intra class distance is also considered as to demonstrate the effectiveness of inter loss.

Convolutional neural network (CNN) is one of the research focuses in machine learning in the last few years. But as the continuous development of CNN in vision and speech, the number of parameters is also increasing, too. CNN, which has millions of parameters, makes the memory of the model very large, and this impedes its widespread especially in mobile device. Based on the observation above, we not only design a CNN pruning method, where we prune unimportant feature maps, but also propose a separability values based number confirmation method which can relatively determine the appropriate pruning number. Experimental results show that, in the cifar-10 dataset, feature maps in each convolutional layer can be pruned by at least 15.6%, up to 59.7%, and the pruning process will not cause any performance loss. We also proved that the confirmation method is effective by a large number of repeated experiments which gradually prune feature maps of each convolutional layer.

The environment perception of road is a key technique for unmanned vehicle. Determining the driving area through segmentation of road image is one of the important methods. The segmentation precisions of the existing methods are not high and some of them are not real-time. To solve these problems, we design a supervised deep Auto-Encoder model to complete the semantic segmentation of road environment image. Firstly, adding a supervised layer to a classical Auto-Encoder, and using the segmentation image of training samples as the supervised information, the model can learn the features useful for segmentation to complete the semantic segmentation. Secondly, the multi-layer stacking method of supervised Auto-Encoder is designed to build the supervised deep Auto-Encoder, because the deep network has more abundant and diversified features. Finally, we verified the method on CamVid. Compare with CNN and FCN, the road segmentation performances such as precision, speed are improved.

In recent years, Wireless Sensor Networks (WSNs) has been successfully applied to the water environment monitoring field. But due to the large area of the monitored waters, the great number of sensor nodes and the vast amount of information collected, the redundancy of data is easy to cause network congestion. In these circumstances, data fusion is essential to WSNs-based water environment monitoring system. Data fusion reduces the energy consumption of communications, but at the same time increases the computational energy consumption. For the purpose of saving energy consumption and prolonging network lifetime, it is necessary and significant to study how to reduce the computation complexity of data fusion. This paper establishes a water environment monitoring network model and a data fusion model in the cluster. On the basis of recursive least squares, the forward and backward recursive algorithms are proposed in order to reduce the computation complexity of data fusion, and the advantages of the new algorithms are analyzed in detail.

Intelligent Traffic Systems (ITS), as integrated systems including control technologies, communication technologies, vehicle sensing and vehicle electronic technologies, have provided valuable solutions to the increasingly serious traffic problems. Hence, in order to achieve efficient management of all types of transportation resources and make better use of ITS, it is necessary and significant to continue study in depth on the architecture and performance of ITS. This paper adopts one kind of stochastic process algebra (SPA)—Performance Evaluation Process Algebra (PEPA) to model and evaluate the process of real-time positioning and route planning in ITS. Meanwhile, the fluid flow approximation is employed to conduct a performance analysis through PEPA models, then the maximize utilization and the throughput of the system can be achieved and analyzed.

Collaborative representation has been successfully used in pattern recognition and machine learning. However, most existing collaborative representation classification methods are to achieve the highest classification accuracy, assuming the same losses for different misclassifications. This assumption may be ineffective in many real-word applications, as misclassification of different types could lead to different losses. Meanwhile, the class distribution of data is highly imbalanced in real-world applications. To address these problems, Cost-sensitive Collaborative Representation based Classification via Probability Estimation Addressing the Class Imbalance Problem method was proposed. The class label of test samples was predict by minimizing the misclassification losses which are obtained via computing the posterior probabilities. In this paper, a Gaussian function was defined as a probability distribution of collaborative representation coefficient vector and it was transformed into collaborative representation framework via logarithmic operator. The experiments on UCI and YaleB databases show that our method performs competitively compared with other methods.

Aerial unmanned vehicle is widely used in many fields, such as weather observation, framing, inspection of infrastructure, monitoring of disaster areas. However, the current aerial unmanned vehicle is difficult to avoid falling in the case of failure. The purpose of this article is to develop an anomaly detection system, which prevents the motor from being used under abnormal temperature conditions, so as to prevent safety flight of the aerial unmanned vehicle. In the anomaly detection system, temperature information of the motor is obtained by DS18B20 sensors. Then, the reinforcement learning, a type of machine learning, is used to determine the temperature is abnormal or not by Raspberrypi processing unit. We also build an user interface to open the screen of Raspberrypi on laptop for observation. In the experiments, the effectiveness of the proposed system to stop the operation state of drone when abnormality exceeds the automatically learned motor temperature. The experimental results demonstrate that the proposed system is possibility for unmanned flight safely by controlling drone from information obtained by attaching temperature sensors.

Underwater optical images are usually influenced by low lighting, high turbidity scattering and wavelength absorption. In order to solve these issues, a great deal of work has been used to improve the quality of underwater images. Most of them used the high-intensity LED for lighting to obtain the high contrast images. However, in high turbidity water, high-intensity LED causes strong scattering and absorption. In this paper, we firstly propose a light field imaging approach for solving underwater depth map estimation problems in low-intensity lighting environment. As a solution, we tackle the problem of de-scattering from light field images by using deep convolutional neural fields in depth estimation. Experimental results show the effectiveness of the proposed method through challenging real world underwater imaging.

To realize the rapid and accurate signature verification, a new image processing method is developed on the basis of the optimal threshold algorithm. Firstly, the improved Gaussian filtering (IGF) algorithm is developed for the signature image to remove the noises. Secondly, the optimal threshold (OT) algorithm is developed to find the optimal threshold for the signature image segmentation. Finally, the deep learning method of a convolutional neural network is used to verify the signature image. It is experimentally proved that the IGF algorithm can get a better filtering effect, and the OT algorithm can obtain the better segmentation result, and the system has the better recognition accuracy.

To eliminating the influence of traveling wave velocity on the fault location accuracy of the power line, a fault location method without wave velocity influence using Wavelet and Clark transform is developed in this paper. On the basis of the reflection characteristic analysis of fault traveling wave, the aerial mode component of the voltage traveling wave is obtained by Clark transform. Then, the fault time when the first three mode maximum appear on the measuring end is determined by the combination of the aerial mode component with Wavelet transform. Furthermore, a set of equations that is composed by the fault time and the fault distance is formed. This method is independent of the wave velocity and is not affected by the traveling wave velocity theoretically. The simulation results show that the developed method can reduce the relative error of the fault distance by less than 5%, which proves the validity and the accuracy of this method.