Browsing by Author "Qiu, Jun-Wei"

Now showing 1 - 6 of 6

A Geometrically Faithful Memetic Algorithm for Searching Sparse Representations of EEG Signals
(2012)
Qiu, Jun-Wei
;
Zao, John K.
;
Chou, Yu-Hsiang
;
資訊工程學系
;
Department of Computer Science
This paper presents a memetic algorithm, christened the natural memetic pursuit (NMP), that was designed to search for sparse signal representations. This algorithm combines global sampling based on a Grassmannian dictionary of atomic signals with local search using natural gradient decent in the signal parameter space. Performance of the algorithm was demonstrated by analyzing Jung and Makig's ERP data sets. It can obtain unbiased sparse representations of EEG signals in far less iteration than its predecessor, the stochastic matching pursuit (SMP).
BLH1-based Collaborative Indoor Localization with Adaptive Multi-lateration and Mobile Encountering
(2016)
Qiu, Jun-Wei
;
Lin, Chien-Pu
;
Tseng, Yu-Chee
;
資訊工程學系
;
Department of Computer Science
The demands on indoor localization services have grown explosively due to increasing popularity of mobile devices. Though various techniques based on wireless signals have been proposed, the new Bluetooth 4.0 and 4.1 technologies introduce both opportunities and challenges for localization. Like all wireless signals, the fluctuations caused by hardware conditions and environmental dynamics may deteriorate the accuracy of localization. Therefore, we propose to use adaptive ranging, which utilizes inter-beacon measurements to "sense" the transient device and environmental conditions, and adjust the parameters of signal propagation model. In this research, we apply this concept to multi-lateration and mobile encountering for localization using Bluetooth advertisements. We also validate that with the proposed adaptive ranging techniques, the impacts of signal fluctuations caused by hardware and environmental conditions can be greatly decreased. Furthermore, we combine the adaptive multi-lateration, encountering mechanism and pedestrian dead reckoning with a particle filter (PF) framework to generate our localization results. Finally, two sets of experiments are performed in a department building. The overall improvement in accuracy is approximately 19.99% using our adaptive localization when comparing to the conventional non-adaptive methods.
Consistent Sparse Representations of EEG ERP and ICA Components Based on Wavelet and Chirplet Dictionaries
(2010)
Qiu, Jun-Wei
;
Zao, John K.
;
Wang, Peng-Hua
;
Chou, Yu-Hsiang
;
資訊工程學系
a randomized search algorithm for sparse representations of EEG event-related potentials (ERPs) and their statistically independent components is presented. This algorithm combines greedy matching pursuit (MP) technique with covariance matrix adaptation evolution strategy (CMA-ES) to select small number of signal atoms from over-complete wavelet and chirplet dictionaries that offer best approximations of quasi-sparse ERP signals. During the search process, adaptive pruning of signal parameters was used to eliminate redundant or degenerative atoms. As a result, the CMA-ES/MP algorithm is capable of producing accurate efficient and consistent sparse representations of ERP signals and their ICA components. This paper explains the working principles of the algorithm and presents the preliminary results of its use.
Continuous Human Location and Posture Tracking by Multiple Depth Sensors
(2014)
Qiu, Jun-Wei
;
Chiang, Ting-Hui
;
Lo, Chi Chung
;
Lin, Li-Min
;
Van, Lan-Da
This paper proposes a continuous location and skeletal tracking system using multiple RGB-Depth sensors (such as Kinects) deployed along a corridor with overlapping coverages. First, we transform the coordinates of all sensor into a unified coordinate. Second, the system recognizes users (such as patients under rehabilitation) from different views of these sensors and classifies them by their patient IDs. Third, the patient information can be continuously handed over among sensors when they move around the area. Experiment results show that the skeletal association during handover achieves approximately 90.61% in accuracy and 96.89% in precision in 44,817 experiment trials. By our observation, injured people possess asymmetric gait parameters especially on the ratio of the duration of the swing/stance phase. For example, the injured foot generally has a longer swinging duration than the healthy side. The proposed system has potential in patient rehabilitation monitoring applications.
M2M Encountering: Collaborative Localization via Instant Inter-Particle Filter Data Fusion
(2016-07-15)
Qiu, Jun-Wei
;
Tseng, Yu-Chee
;
資訊工程學系
;
Department of Computer Science
Much research has addressed indoor localization by integrating portable/wearable sensing and communication technologies. While GPS has dominated the outdoor environments, indoor localization schemes have to consider different observations, such as radio-frequency signals, vision or motions data, and apply data fusion to combine various sensor inputs. In this paper, we observe that when two devices meet up, which we call machine-to-machine (M2M) encountering, they can collaboratively calibrate each other\'s potential locations via M2M communications. We apply this technique to particle filter (PF), a common fusion technique, and show how to take the M2M encountering opportunities, which may happen frequently in crowded areas, to allow user devices to collaboratively calibrate their locations. Hence, the PF technique, which normally fuses observations from individual devices, is extended to an inter-PF, cross-device domain. We validate our inter-PF solution by simulations as well as a prototype system with smartphones and ZigBee infrastructure deployed in an office building. It is verified that the inter-PF helps converge the positioning results more rapidly, and improves location quality. Also, the proposed encountering mechanism has potential to be applied to other localization algorithms to improve their accuracy.
Wearable Localization by Particle Filter with the Assistance of Inertial and Visual Sensors
(2014-01-01)
Huang, Sz-Pin
;
Qiu, Jun-Wei
;
Lo, Chj-Chung
;
Tseng, Yu-Chee
;
資訊工程學系
Indoor positioning has been intensively studied recently due to the exploding demands of indoor mobile applications. While numerous works have employed wireless signals or dead-reckoning techniques, wearable computing poses new opportunities as well as challenges to the localization problem. This research studies the wearable localization problem by proposing a particle filter-based scheme to fuse the inputs from wearable inertial and visual sensors on human body. Specifically, the filter takes inertial measurements, wireless signals, visual landmarks, and indoor floor plans as inputs for location tracking. The inertial signals imply human body movements, the wireless signals indicate a rough absolute region inside a building, while the visual landmarks provide relative angles viewed from particular positions to these markers. Furthermore, a head-mounted display provides intuitive and friendly interfaces to users. The proposed system has also been prototyped and tested in our campus, and the experiments demonstrate an average localization error of about one meter.