Search

This paper proposes a novel indoor device-free localization and tracking approach using the received signal-strength indicators (RSSIs) of WiFi signals. The RSSI feature-vectors simulated by a channel-propagation emulator software are adopted as the training data for our proposed scheme. Prevalent discriminative machine-learning methods are used to predict the locations of a moving human-object. Hidden Markov models (HMMs) are also incorporated with such machine-learning techniques for robust and reliable indoor tracking. In this work, we partition the given indoor geometry into several equi-sized zones and then convert the underlying localization/tracking problem to the classical multi-classification problem. Simulation results demonstrate that the gradient boosting decision-tree (GBDT) classifier in conjunction with the Viterbi algorithm over hidden Markov models leads to the highest localization-accuracies of 83.9% for eight zones and 71.4% for sixteen zones. As a result, our proposed new indoor localization and tracking scheme can be very promising for many indoor device-free surveillance applications in the future.

This paper introduces a new indoor-localization approach using a deep-learning network for which the received signal-strength indicator (RSSI) is adopted as the radiofrequency fingerprint. In our proposed scheme, the RSSIs which are estimated by a channel-propagation emulator software, are adopted as the input features for deep learning. This approach is measurement-free and thus it is very cost-effective and convenient to users. A multilayer perceptron (MLP) is constructed to predict the location(s) of the mobile object(s). The time evolution of the predicted locations of an object will form the predicted trajectory thereby. Because deep-learning networks require tremendous training data to achieve good prediction accuracy, we propose to partition the indoor geometry of interest (ex., a room) into several zones. Preliminary simulation results demonstrate that the AUC (area under the receiver-operating characteristic curve) can reach up to 0.89 for a room partitioned into eight zones. Our proposed new indoor-localization scheme in this work can be a rare but promising localization technology, which is neither passive nor active as other existing prevalent localization methods.

Nowadays, automatic object-size measurement (AOSM) techniques are in high demand since there are numerous applications including three-dimensional multimedia entertainment games, cyber-physical interaction, and real-time geographical survey. When a scene is complex with many objects, the AOSM problem is quite challenging because it is very hard to describe the object of interest a priori. In this paper, we present a novel efficient automatic human-height measurement scheme using a single camera. Our proposed new scheme is very robust so that the heights of multiple persons within the scene can be estimated automatically and simultaneously. The experimental results demonstrate that the average error percentage is less than 2% in general.

In this paper, we propose a new hierarchical radio-frequency identification (RFID) tag- recognition method based on the graph-based automatic modulation classification (AMC) and direct-sequence spread-spectrum (DSSS). In our proposed method, RFID tags can be modulated using different modulation schemes according to different scenarios (e.g., different users or different devices). For each modulation scheme, the direct-sequence spread-spectrum strategy is employed to allow simultaneous transmissions of multiple tags. In the first hierarchy, different modulation types are employed to distinguish different users, the graph-based AMC is built upon the periodicity of the modulated signals. The cyclic spectrum of the received signal is first established. The graph representation is then constructed according to the cyclic spectrum. Ultimately, robust features are extracted from the graph representation. In the second hierarchy, the DSSS scheme is utilized to differentiate the control or sensed data carried by individual tags, the signature sequence set with low cross- correlations is generated from Gold and Kasami sequences. In the third hierarchy, the information data sequences are thus spread by these signature sequences. In our proposed new RFID framework, multiple tags can transmit the signals simultaneously in the same frequency band where the dominant tag signal will be identified and its carried information can be recovered. Monte Carlo simulation results demonstrate the promising performance of our proposed new scheme.

The electronic media, such as digital magazines and newspapers, are becoming more and more popular nowadays. The computer-aided information retrieval of such electronic documents is crucial for any further data-mining applications. In this paper, we proposed a novel electronic scissoring algorithm to automatically extract the image content from an arbitrary electronic document. Our proposed new algorithm contains two steps, namely the information-content extraction and the image-content detection. In the first step, everything other than the background is extracted using a computationally-efficient variance filter. The information content usually consists of text, image, and delimiters. The next step, the entropy measure is employed to distinguish the images from other content. Experiments are carried out using data captured from random web-pages. The receiver-operating characteristic curves have been presented to demonstrate the effectiveness of our proposed new algorithm.

Quick Response (QR) code is a two-dimensional bar-code, which is quite popular due to its excellent storage capacity and error resilience. Generally, QR codes are widely used to store text information such as URL links, geographical coordinates, name cards, inventory information, authorship, etc. To reach the current limit of QR codes, in this paper, we would like to propose an innovative multimedia archiving technique, which is built upon the advanced signal processing scheme to tackle the multiple QR codes all at once. The multimedia data often include texts, images, audio data, special-purpose codes, etc. The recently proposed software-defined multiplexing code (SDMC) can be applied to combine all of them even though each type of data would have the individual data format different from others. Our proposed new archiving technique involves two phases, namely multimedia-amalgamation (MA) and multimedia-detachment (MD). In the MA phase, the multimedia data, regardless of their data format, can be converted to the binary streams; then the SDMC will be employed to aggregate them together in a longer binary stream; such an ultimate long binary stream will be converted to the QR codes. The QR codes, protected by the inherent error-correction mechanism, will thus be placed on a sheet or multiple sheets with a uniform spacing. The sheet(s) containing multiple QR codes can thus be archived in soft (PDF) or hard (print-out) copies. In the MD (recovery) phase, one can scan these PDF files and employ our designed signal processing algorithms to separate the individual QR codes; then the corresponding QR decoder can convert these QR codes back to the original binary stream. Finally the unstuffing algorithm in the SDMC can detach the individual data from the long binary stream composed by the multimedia mixture.

Nowadays, the sufficient quality-of-service (QoS) provision for mobile applications remains a major challenge in any wireless network. Conventional sufficient QoS provision techniques using resource allocation, data scheduling, and cross-layer optimization have been proposed to tackle this problem. Nevertheless, due to the unpredictable nature of wireless channel conditions, the QoS improvements resulting from the aforementioned approaches are often unsatisfactory. In this paper, we would like to follow our previous concept, namely user-placement ushering (UPU), by use of the user's mobility. A mobile user can be relocated to an optimal, or at least a better place to boost the QoS for a particular application. Novel fast algorithms are devised here to usher (guide) the mobile user to an appropriate spot with sufficient QoS, which is nearby. We address the UPU problem to accommodate realistic complex indoor environments, where obstacles are present. Our simulation results have demonstrated that our new fast UPU algorithms are capable of finding new appropriate locations satisfying the required QoSs for different wireless network applications.

This paper proposes a novel approach to measure the object size using a regular digital camera. Nowadays, the remote object-size measurement is very crucial to many multimedia applications. Our proposed computer-aided automatic object-size measurement technique is based on a new depth-information extraction (range finding) scheme using a regular digital camera. The conventional range finders are often carried out using the passive method such as stereo cameras or the active method such as ultrasonic and infrared equipment. They either require the cumbersome set-up or deal with point targets only. The proposed approach requires only a digital camera with certain image processing techniques and relies on the basic principles of visible light. Experiments are conducted to evaluate the performance of our proposed new object-size measurement mechanism. The average error-percentage of this method is below 2%. It demonstrates the striking effectiveness of our proposed new method.

Automatic size measurements have been intriguing lately since they have a lot of applications such as cyber-physical interaction, three-dimensional multimedia entertainment games, navigation and target tracking, real-time surveying of the environment using artificial intelligence. Conventional accurate approaches are mostly based on the active techniques, including laser-beams, ultrasonics, radio-frequency transceivers, etc. These existing approaches can recognize neither the object nor the context of the scene; thus advanced image processing methods using visible-light digital cameras are more preferable on the other hand. In this paper, we utilize the basic visible-light propagation principles to build a novel efficient automatic size measurement algorithm, which can be employed by one or two digital cameras. The experimental results show that less than 3% of striking accuracy can be achieved on average using our proposed new method.

In this paper, a novel approach to locate the activation points of atrial fibrillation (AF) is proposed. This new method is built upon machine learning, where common parameters, such as dominant frequency, first harmonic frequency, etc., are adopted. Features are extracted from the original electrocardiography (ECG) and then clustering is performed to classify the ECG signals into two groups, namely activation and nonactivation points. The experimental results are compared with those from the state-of-the-art system, Topera, used in East Jefferson General Hospital nowadays.