Analyzing Performance of YOLOx for Detecting Vehicles in Bad Weather Conditions.

Recent advancements in computer vision technology, developments in sensors and sensor-collecting approaches, and the use of deep and transfer learning approaches have excelled in the development of autonomous vehicles. On-road vehicle detection has become a task of significant importance, especially due to exponentially increasing research on autonomous vehicles during the past few years. With high-end computing resources, a large number of deep learning models have been trained and tested for on-road vehicle detection recently. Vehicle detection may become a challenging process especially due to varying light and weather conditions like night, snow, sand, rain, foggy conditions, etc. In addition, vehicle detection should be fast enough to work in real time. This study investigates the use of the recent YOLO version, YOLOx, to detect vehicles in bad weather conditions including rain, fog, snow, and sandstorms. The model is tested on the publicly available benchmark dataset DAWN containing images containing four bad weather conditions, different illuminations, background, and number of vehicles in a frame. The efficacy of the model is evaluated in terms of precision, recall, and mAP. The results exhibit the better performance of YOLOx-s over YOLOx-m and YOLOx-l variants. YOLOx-s has 0.8983 and 0.8656 mAP for snow and sandstorms, respectively, while its mAP for rain and fog is 0.9509 and 0.9524, respectively. The performance of models is better for snow and foggy weather than rainy weather sandstorms. Further experiments indicate that enhancing image quality using multiscale retinex improves YOLOx performance.

Smartphone Sensors for Indoor Positioning.

The explosive growth and wide proliferation of mobile devices, the majority of which are smartphones, led to the inception of several novel and intuitive services, including on-the-go services, online customer services, and location-based services (LBS) [...].

Nested Bee Hive: A Conceptual Multilayer Architecture for 6G in Futuristic Sustainable Smart Cities.

Several smart city ideas are introduced to manage various problems caused by overpopulation, but the futuristic smart city is a concept based on dense and artificial-intelligence-centric cities. Thus, massive device connectivity with huge data traffic is expected in the future where communication networks are expected to provide ubiquity, high quality of service, and on-demand content for a large number of interconnected devices. The sixth-generation (6G) network is considered the problem-solving network of futuristic cities, with huge bandwidth and low latency. The expected 6G of the radio access network is based on terahertz (THz) waves with the capability of carrying up to one terabit per second (Tbps). THz waves have the capability of carrying a large amount of data but these waves have several drawbacks, such as short-range and atmospheric attenuation. Hence, these problems can introduce complications and hamper the performance of the 6G network. This study envisions futuristic smart cities using 6G and proposes a conceptual terrestrial network (TN) architecture for 6G. The nested Bee Hive is a scalable multilayer architecture designed to meet the needs of futuristic smart cities. Moreover, we designed the multilayer network infrastructure while considering the expectations from a network of futuristic smart cities and the complications of THz waves. Extensive simulations are performed using different pathfinding algorithms in the 3D multilayer domain to evaluate the performance of the proposed architecture and set the dynamics of futuristic communication of 6G.

Empirical Overview of Benchmark Datasets for Geomagnetic Field-Based Indoor Positioning.

Indoor positioning and localization have been regarded as some of the most widely researched areas during the last decade. The wide proliferation of smartphones and the availability of fast-speed internet have initiated several location-based services. Concerning the importance of precise location information, many sensors are embedded into modern smartphones. Besides Wi-Fi positioning, a rich variety of technologies have been introduced or adopted for indoor positioning such as ultrawideband, infrared, radio frequency identification, Bluetooth beacons, pedestrian dead reckoning, and magnetic field, etc. However, special emphasis is put on infrastructureless approaches like Wi-Fi and magnetic field-based positioning, as they do not require additional infrastructure. Magnetic field positioning is an attractive solution for indoors; yet lack of public benchmarks and selection of suitable benchmarks are among the big challenges. While several benchmarks have been introduced over time, the selection criteria of a benchmark are not properly defined, which leads to positioning results that lack generalization. This study aims at analyzing various public benchmarks for magnetic field positioning and highlights their pros and cons for evaluation positioning algorithms. The concept of DUST (device, user, space, time) and DOWTS (dynamicity, orientation, walk, trajectory, and sensor fusion) is introduced which divides the characteristics of the magnetic field dataset into basic and advanced groups and discusses the publicly available datasets accordingly.

Enhancing Performance of Magnetic Field Based Indoor Localization Using Magnetic Patterns from Multiple Smartphones.

Wide expansion of smartphones triggered a rapid demand for precise localization that can meet the requirements of location-based services. Although the global positioning system is widely used for outdoor positioning, it cannot provide the same accuracy for the indoor. As a result, many alternative indoor positioning technologies like Wi-Fi, Bluetooth Low Energy (BLE), and geomagnetic field localization have been investigated during the last few years. Today smartphones possess a rich variety of embedded sensors like accelerometer, gyroscope, and magnetometer that can facilitate estimating the current location of the user. Traditional geomagnetic field-based fingerprint localization, although it shows promising results, it is limited by the fact that various smartphones have embedded magnetic sensors from different manufacturers and the magnetic field strength that is measured from these smartphones vary significantly. Consequently, the localization performance from various smartphones is different even when the same localization approach is used. So devising an approach that can provide similar performance with various smartphones is a big challenge. Contrary to previous works that build the fingerprint database from the geomagnetic field data of a single smartphone, this study proposes using the geomagnetic field data collected from multiple smartphones to make the geomagnetic field pattern (MP) database. Many experiments are carried out to analyze the performance of the proposed approach with various smartphones. Additionally, a lightweight threshold technique is proposed that can detect user motion using the acceleration data. Results demonstrate that the localization performance for four different smartphones is almost identical when tested with the database made using the magnetic field data from multiple smartphones than that of which considers the magnetic field data from only one smartphone. Moreover, the performance comparison with previous research indicates that the overall performance of smartphones is improved.

DeepLocate: Smartphone Based Indoor Localization with a Deep Neural Network Ensemble Classifier.

A quickly growing location-based services area has led to increased demand for indoor positioning and localization. Undoubtedly, Wi-Fi fingerprint-based localization is one of the promising indoor localization techniques, yet the variation of received signal strength is a major problem for accurate localization. Magnetic field-based localization has emerged as a new player and proved a potential indoor localization technology. However, one of its major limitations is degradation in localization accuracy when various smartphones are used. The localization performance is different from various smartphones even with the same localization technique. This research leverages the use of a deep neural network-based ensemble classifier to perform indoor localization with heterogeneous devices. The chief aim is to devise an approach that can achieve a similar localization accuracy using various smartphones. Features extracted from magnetic data of Galaxy S8 are fed into neural networks (NNs) for training. The experiments are performed with Galaxy S8, LG G6, LG G7, and Galaxy A8 smartphones to investigate the impact of device dependence on localization accuracy. Results demonstrate that NNs can play a significant role in mitigating the impact of device heterogeneity and increasing indoor localization accuracy. The proposed approach is able to achieve a localization accuracy of 2.64 m at 50% on four different devices. The mean error is 2.23 m, 2.52 m, 2.59 m, and 2.78 m for Galaxy S8, LG G6, LG G7, and Galaxy A8, respectively. Experiments on a publicly available magnetic dataset of Sony Xperia M2 using the proposed approach show a mean error of 2.84 m with a standard deviation of 2.24 m, while the error at 50% is 2.33 m. Furthermore, the impact of devices on various attitudes on the localization accuracy is investigated.

Catastrophic factors involved in road accidents: Underlying causes and descriptive analysis.

South Korea is ranked as 4th among 34 nations of the Organization for Economic Cooperation and Development with 102 deaths in road accidents per one million population. This paper aims to investigate the factors associated with road accidents in South Korea. The rainfall data of the Korea Meteorological Administration and road accidents data of Traffic Accident Analysis System of Korea Road Traffic Authority is analyzed for this purpose. In this connection, multivariate regression analysis and ratio analysis with the descriptive analysis are performed to uncover the catastrophic factors involved. In turn, the results reveal that traffic volume is the leading factor in road accidents. The limited road extension of 1.47% compared to the 4.14% per annum growth of the vehicles is resulting in road accidents at such a large scale. The increasing proportion of passenger cars accelerate road accidents as well. 56% of accidents occur by the infringement of safety driving violations. The drivers with higher driving experience tend to have a higher accident ratio. The collected data is analyzed in terms of gender, driver experience, type of violations and accidents as well as the associated time of the accidents when they happen. The results indicate that 36.29% and 53.01% of accidents happen by male drivers in the day and night time, respectively. 29.15% of crashes happen due to safety infringement and violations of 41 to 60 years old drivers. The results demonstrate that population density is associated with the accidents frequency and lower density results in an increased number of accidents. The necessity of the state-of-the-art regulations to govern the urban road traffic is beyond dispute, and it becomes even more crucial for citizens' relief since in our daily lives road accidents are getting more diverse.

Indoor Positioning on Disparate Commercial Smartphones Using Wi-Fi Access Points Coverage Area.

The applications of location-based services require precise location information of a user both indoors and outdoors. Global positioning system's reduced accuracy for indoor environments necessitated the initiation of Indoor Positioning Systems (IPSs). However, the development of an IPS which can determine the user's position with heterogeneous smartphones in the same fashion is a challenging problem. The performance of Wi-Fi fingerprinting-based IPSs is degraded by many factors including shadowing, absorption, and interference caused by obstacles, human mobility, and body loss. Moreover, the use of various smartphones and different orientations of the very same smartphone can limit its positioning accuracy as well. As Wi-Fi fingerprinting is based on Received Signal Strength (RSS) vector, it is prone to dynamic intrinsic limitations of radio propagation, including changes over time, and far away locations having similar RSS vector. This article presents a Wi-Fi fingerprinting approach that exploits Wi-Fi Access Points (APs) coverage area and does not utilize the RSS vector. Using the concepts of APs coverage area uniqueness and coverage area overlap, the proposed approach calculates the user's current position with the help of APs' intersection area. The experimental results demonstrate that the device dependency can be mitigated by making the fingerprinting database with the proposed approach. The experiments performed at a public place proves that positioning accuracy can also be increased because the proposed approach performs well in dynamic environments with human mobility. The impact of human body loss is studied as well.

Floor Identification Using Magnetic Field Data With Smartphone Sensors.

Floor identification plays a key role in multi-story indoor positioning and localization systems. Current floor identification systems rely primarily on Wi-Fi signals and barometric pressure data. Barometric systems require installation of additional standalone sensors to perform floor identification. Wi-Fi systems, on the other hand, are vulnerable to the dynamic environment and adverse effects of path loss, shadowing, and multipath fading. In this paper, we take advantage of a pervasive magnetic field to compensate for the limitations of these systems. We employ smartphone sensors to make the proposed scheme infrastructure free and cost-effective. We use smartphone magnetic sensors to identify the floors in a multi-story building with improved accuracy. Floor identification is performed with user activities of normal walking, call listening, and phone swinging. Various machine learning techniques are leveraged to identify user activities. Extensive experiments are performed to evaluate the proposed magnetic-data-based floor identification scheme. Additionally, the impact of device heterogeneity on floor identification is investigated using Samsung Galaxy S8, LG G6, and LG G7 smartphones. Research results demonstrate that the magnetic floor identification outperforms barometric and Wi-Fi-enabled floor detection techniques. A floor change module is incorporated to further enhance the accuracy of floor identification.

Wi-Fi-Based Effortless Indoor Positioning System Using IoT Sensors.

Wi-Fi positioning based on fingerprinting has been considered as the most widely used technology in the field of indoor positioning. The fingerprinting database has been used as an essential part of the Wi-Fi positioning system. However, the offline phase of the calibration involves a laborious task of site analysis which involves costs and a waste of time. We offer an indoor positioning system based on the automatic generation of radio maps of the indoor environment. The proposed system does not require any effort and uses Wi-Fi compatible Internet-of-Things (IoT) sensors. Propagation loss parameters are automatically estimated from the online feedback of deployed sensors and the radio maps are updated periodically without any physical intervention. The proposed system leverages the raster maps of an environment with the wall information only, against computationally extensive techniques based on vector maps that require precise information on the length and angles of each wall. Experimental results show that the proposed system has achieved an average accuracy of 2 m, which is comparable to the survey-based Wi-Fi fingerprinting technique.

Suitable Combination of Direct Intensity Modulation and Spreading Sequence for LIDAR with Pulse Coding.

In the coded pulse scanning light detection and ranging (LIDAR) system, the number of laser pulses used at a given measurement point changes depending on the modulation and the method of spreading used in optical code-division multiple access (OCDMA). The number of laser pulses determines the pulse width, output power, and duration of the pulse transmission of a measurement point. These parameters determine the maximum measurement distance of the LIDAR and the number of measurement points that can be employed per second. In this paper, we suggest possible combinations of modulation and spreading technology that can be used for OCDMA, evaluate their performance and characteristics, and study optimal combinations according to varying operating environments.

BLocate: A Building Identification Scheme in GPS Denied Environments Using Smartphone Sensors.

Indoor localization systems assume that the user's current building is known by the GPS (Global Positioning System). However, such assumptions do not hold true in GPS denied environments or where the GPS cannot determine the user's definite location. We present a novel solution to identify the building where the user is present now. The proposed building identification method works on the pervasive magnetic field using a smartphone. The accelerometer data determines the user's activity of being stationary or walking. An Artificial Neural Network is used to identify the user's activities and it shows good results. The magnetometer data is used to identify the user's current building using the fingerprinting approach. Contrary to a traditional fingerprinting approach which stores intensity values, we utilize the patterns formed by the magnetic field strength in the form of a Binary Grid (BG). The BG approach overcomes the limitation of Dynamic Time Warping (DTW) whose performance is degraded when the magnitude of the magnetic data is changed. The experiments are performed with Samsung Galaxy S8 for eight various buildings with different altitudes and number of floors in Yeungnam University, Korea. The results demonstrate that the proposed building identification method can potentially be deployed for building identification. The precision, UAR (Unweighted Average Recall), F score, and Cohen's Kappa values are used to determine the performance of the proposed system. The proposed systems shows very promising results. The system operates without any aid from any infrastructure dependent technologies like GPS or WiFi. Furthermore, we performed many experiments to investigate the impact of isolated points data to build fingerprint database on system's accuracy with 1 m and 2 m distance. Results illustrate that by trading off a minor accuracy, survey labor can be reduced by 50 percent.

MagIO: Magnetic Field Strength Based Indoor- Outdoor Detection with a Commercial Smartphone.

A wide range of localization techniques has been proposed recently that leverage smartphone sensors. Context awareness serves as the backbone of these localization techniques, which helps them to shift the localization technologies to improve efficiency and energy utilization. Indoor-outdoor (IO) context sensing plays a vital role for such systems, which serve both indoor and outdoor localization. IO systems work with collaborative technologies including the Global Positioning System (GPS), cellular tower signals, Wi-Fi, Bluetooth and a variety of smartphone sensors. GPS- and Wi-Fi-based systems are power hungry, and their accuracy is severed by limiting factors like multipath, shadowing, etc. On the other hand, various built-in smartphone sensors can be deployed for environmental sensing. Although these sensors can play a crucial role, yet they are very less studied. This research aims at investigating the use of ambient magnetic field data alone from a smartphone for IO detection. The research first investigates the feasibility of utilizing magnetic field data alone for IO detection and then extracts different features suitable for IO detection to be used in machine learning-based classifiers to discriminate between indoor and outdoor environments. The experiments are performed at three different places including a subway station, a shopping mall and Yeungnam University (YU), Korea. The training data are collected from one spot of the campus, and testing is performed with data from various locations of the above-mentioned places. The experiment involves Samsung Galaxy S8, LG G6 and Samsung Galaxy Round smartphones. The results show that the magnetic data from smartphone magnetic sensor embody enough information and can discriminate the indoor environment from the outdoor environment. Naive Bayes (NB) outperforms with a classification accuracy of 83.26%, as against Support vector machines (SVM), random induction (RI), gradient boosting machines (GBM), random forest (RF), k-nearest neighbor (kNN) and decision trees (DT), whose accuracies are 67.21%, 73.38%, 73.40%, 78.59%, 69.53% and 68.60%, respectively. kNN, SVM and DT do not perform well when noisy data are used for classification. Additionally, other dynamic scenarios affect the attitude of magnetic data and degrade the performance of SVM, RI and GBM. NB and RF prove to be more noise tolerant and environment adaptable and perform very well in dynamic scenarios. Keeping in view the performance of these classifiers, an ensemble-based stacking scheme is presented, which utilizes DT and RI as the base learners and naive Bayes as the ensemble classifier. This approach is able to achieve an accuracy of 85.30% using the magnetic data of the smartphone magnetic sensor. Moreover, with an increase in training data, the accuracy of the stacking scheme can be elevated by 0.83%. The performance of the proposed approach is compared with GPS-, Wi-Fi- and light sensor-based IO detection.

Independent Biaxial Scanning Light Detection and Ranging System Based on Coded Laser Pulses without Idle Listening Time.

The goal of light detection and ranging (LIDAR) systems is to achieve high-resolution three-dimensional distance images with high refresh rates and long distances. In scanning LIDAR systems, an idle listening time between pulse transmission and reception is a significant obstacle to accomplishing this goal. We apply intensity-modulated direct detection (IM/DD) optical code division multiple access (OCDMA) using nonreturn-to-zero on-off keying to eliminate the idle listening time in scanning LIDAR systems. The transmitter records time information while emitting a coded laser pulse in the measurement angle derived from the pixel information as the measurement direction. The receiver extracts and decodes the reflected laser pulses and estimates the distance to the target using time-of-flight until the pulse is received after being transmitted. Also, we rely on a series of pulses and eliminate alien pulses via several detection decision steps to enhance the robustness of the decision result. We built a prototype system and evaluated its performance by measuring black matte and white paper walls and assessing object detection by measuring a watering can in front of the black matte paper wall. This LIDAR system eliminated both shot and background noises in the reception process and measured greater distances with improvements in accuracy and precision.

mPILOT-Magnetic Field Strength Based Pedestrian Indoor Localization.

An indoor localization system based on off-the-shelf smartphone sensors is presented which employs the magnetometer to find user location. Further assisted by the accelerometer and gyroscope, the proposed system is able to locate the user without any prior knowledge of user initial position. The system exploits the fingerprint database approach for localization. Traditional fingerprinting technology stores data intensity values in database such as RSSI (Received Signal Strength Indicator) values in the case of WiFi fingerprinting and magnetic flux intensity values in the case of geomagnetic fingerprinting. The down side is the need to update the database periodically and device heterogeneity. We solve this problem by using the fingerprint database of patterns formed by magnetic flux intensity values. The pattern matching approach solves the problem of device heterogeneity and the algorithm's performance with Samsung Galaxy S8 and LG G6 is comparable. A deep learning based artificial neural network is adopted to identify the user state of walking and stationary and its accuracy is 95%. The localization is totally infrastructure independent and does not require any other technology to constraint the search space. The experiments are performed to determine the accuracy in three buildings of Yeungnam University, Republic of Korea with different path lengths and path geometry. The results demonstrate that the error is 2⁻3 m for 50 percentile with various buildings. Even though many locations in the same building exhibit very similar magnetic attitude, the algorithm achieves an accuracy of 4 m for 75 percentile irrespective of the device used for localization.

LOCALI: Calibration-Free Systematic Localization Approach for Indoor Positioning.

Recent advancements in indoor positioning systems are based on infrastructure-free solutions, aimed at improving the location accuracy in complex indoor environments without the use of specialized resources. A popular infrastructure-free solution for indoor positioning is a calibration-based positioning, commonly known as fingerprinting. Fingerprinting solutions require extensive and error-free surveys of environments to build radio-map databases, which play a key role in position estimation. Fingerprinting also requires random updates of the database, when there are significant changes in the environment or a decrease in the accuracy. The calibration of the fingerprinting database is a time-consuming and laborious effort that prevents the extensive adoption of this technique. In this paper, we present a systematic LOCALIzation approach, "LOCALI", for indoor positioning, which does not require a calibration database and extensive updates. The LOCALI exploits the floor plan/wall map of the environment to estimate the target position by generating radio maps by integrating path-losses over certain trajectories in complex indoor environments, where triangulation using time information or the received signal strength level is highly erroneous due to the fading effects caused by multi-path propagation or absorption by environmental elements or varying antenna alignment. Experimental results demonstrate that by using the map information and environmental parameters, a significant level of accuracy in indoor positioning can be achieved. Moreover, this process requires considerably lesser effort compared to the calibration-based techniques.

A non-invasive approach to investigation of ventricular blood pressure using cardiac sound features.

Heart sounds (HSs) are produced by the interaction of the heart valves, great vessels, and heart wall with blood flow. Previous researchers have demonstrated that blood pressure can be predicted by exploring the features of cardiac sounds. These features include the amplitude of the HSs, the ratio of the amplitude, the systolic time interval, and the spectrum of the HSs. A single feature or combinations of several features have been used for prediction of blood pressure with moderate accuracy. Experiments were conducted with three beagles under various levels of blood pressure induced by different doses of epinephrine. The HSs, blood pressure in the left ventricle and electrocardiograph signals were simultaneously recorded. A total of 31 records (18 262 cardiac beats) were collected. In this paper, 91 features in various domains are extracted and their linear correlations with the measured blood pressures are examined. These features are divided into four groups and applied individually at the input of a neural network to predict the left ventricular blood pressure (LVBP). The analysis shows that non-spectral features can track changes of the LVBP with lower standard deviation. Consequently, the non-spectral feature set gives the best prediction accuracy. The average correlation coefficient between the measured and the predicted blood pressure is 0.92 and the mean absolute error is 6.86 mmHg, even when the systolic blood pressure varies in the large range from 90 mmHg to 282 mmHg. Hence, systolic blood pressure can be accurately predicted even when using fewer HS features. This technique can be used as an alternative to real-time blood pressure monitoring and it has promising applications in home health care environments.

LIDAR pulse coding for high resolution range imaging at improved refresh rate.

In this study, a light detection and ranging system (LIDAR) was designed that codes pixel location information in its laser pulses using the direct- sequence optical code division multiple access (DS-OCDMA) method in conjunction with a scanning-based microelectromechanical system (MEMS) mirror. This LIDAR can constantly measure the distance without idle listening time for the return of reflected waves because its laser pulses include pixel location information encoded by applying the DS-OCDMA. Therefore, this emits in each bearing direction without waiting for the reflected wave to return. The MEMS mirror is used to deflect and steer the coded laser pulses in the desired bearing direction. The receiver digitizes the received reflected pulses using a low-temperature-grown (LTG) indium gallium arsenide (InGaAs) based photoconductive antenna (PCA) and the time-to-digital converter (TDC) and demodulates them using the DS-OCDMA. When all of the reflected waves corresponding to the pixels forming a range image are received, the proposed LIDAR generates a point cloud based on the time-of-flight (ToF) of each reflected wave. The results of simulations performed on the proposed LIDAR are compared with simulations of existing LIDARs.

Phonocardiogram signal compression using sound repetition and vector quantization.

BACKGROUND: A phonocardiogram (PCG) signal can be recorded for long-term heart monitoring. A huge amount of data is produced if the time of a recording is as long as days or weeks. It is necessary to compress the PCG signal to reduce storage space in a record and play system. In another situation, the PCG signal is transmitted to a remote health care center for automatic analysis in telemedicine. Compression of the PCG signal in that situation is necessary as a means for reducing the amount of data to be transmitted. Since heart beats are of a cyclical nature, compression can make use of the similarities in adjacent cycles by eliminating repetitive elements as redundant. This study proposes a new compression method that takes advantage of these repetitions. METHODS: Data compression proceeds in two stages, a training stage followed by the compression as such. In the training stage, a section of the PCG signal is selected and its sounds and murmurs (if any) decomposed into time-frequency components. Basic components are extracted from these by clustering and collected to form a dictionary that allows the generative reconstruction and retrieval of any heart sound or murmur. In the compression stage, the heart sounds and murmurs are reconstructed from the basic components stored in the dictionary. Compression is made possible because only the times of occurrence and the dictionary indices of the basic components need to be stored, which greatly reduces the number of bits required to represent heart sounds and murmurs. The residual that cannot be reconstructed in this manner appears as a random sequence and is further compressed by vector quantization. What we propose are quick search parameters for this vector quantization. RESULTS: For normal PCG signals the compression ratio ranges from 20 to 149, for signals with median murmurs it ranges from 14 to 35, and for those with heavy murmurs, from 8 to 20, subject to a degree of distortion of ~5% (in percent root-mean-square difference) and a sampling frequency of 4kHz. DISCUSSION: We discuss the selection of the training signal and the contribution of vector quantization. Performance comparisons between the method proposed in this study and existing methods are conducted by computer simulations. CONCLUSIONS: When recording and compressing cyclical sounds, any repetitive components can be removed as redundant. The redundancies in the residual can be reduced by vector quantization. The method proposed in this study achieves a better performance than existing methods.