Satu-Marja Mäkelä

Identifying users of portable devices from gait pattern with accelerometers

Identifying users of portable devices from gait signals acquired with three-dimensional accelerometers was studied. Three approaches, correlation, frequency domain and data distribution statistics, were used. Test subjects (N=36) walked with fast, normal and slow walking speeds in enrolment and test sessions on separate days wearing the accelerometer device on their belt, at back. It was shown to be possible to identify users with this novel gait recognition method. Best equal error rate (EER=7%) was achieved with the signal correlation method, while the frequency domain method and two variations of the data distribution statistics method produced EER of 10%, 18% and 19%, respectively.

Identifying people from gait pattern with accelerometers

Protecting portable devices is becoming more important, not only because of the value of the devices themselves, but for the value of the data in them and their capability for transactions, including m-commerce and m-banking. An unobtrusive and natural method for identifying the carrier of portable devices is presented. The method uses acceleration signals produced by sensors embedded in the portable device. When the user carries the device, the acceleration signal is compared with the stored template signal. The method consists of finding individual steps, normalizing and averaging them, aligning them with the template and computing cross-correlation, which is used as a measure of similarity. Equal Error Rate of 6.4% is achieved in tentative experiments with 36 test subjects.

Bayesian approach to sensor-based context awareness

AbstractThe usability of a mobile device and services can be enhanced by context awareness. The aim of this experiment was to expand the set of generally recognizable constituents of context concerning personal mobile device usage. Naive Bayesian networks were applied to classify the contexts of a mobile device user in her normal daily activities. The distinguishing feature of this experiment in comparison to earlier context recognition research is the use of a naive Bayes framework, and an extensive set of audio features derived partly from the algorithms of the upcoming MPEG-7 standard. The classification was based mainly on audio features measured in a home scenario. The classification results indicate that with a resolution of one second in segments of 5–30 seconds, situations can be extracted fairly well, but most of the contexts are likely to be valid only in a restricted scenario. Naive Bayes framework is feasible for context recognition. In real world conditions, the recognition accuracy using leave-one-out cross validation was 87% of true positives and 95% of true negatives, averaged over nine eight-minute scenarios containing 17 segments of different lengths and nine different contexts. Respectively, the reference accuracies measured by testing with training data were 88% and 95%, suggesting that the model was capable of covering the variability introduced in the data on purpose. Reference recognition accuracy in controlled conditions was 96% and 100%, respectively. However, from the applicability viewpoint, generalization remains a problem, as from a wider perspective almost any feature may refer to many possible real world situations.

Soft biometrics-combining body weight and fat measurements with fingerprint biometrics

The aim of this study was to examine whether using soft biometrics, i.e. easily measurable personal characteristics, such as weight and fat percentage, can improve the performance of biometrics in verification type applications. Fusing fingerprint biometrics with soft biometrics, in this case body weight measurements, decreased the total error rate (TER) from 3.9% to 1.5% in an experiment with 62 test subjects. This result shows that simple physiological measurements can be used to support biometric recognition. Furthermore, soft biometrics are unobtrusive, there is no risk of identity theft, the perception of the big-brother effect is small, the equipment needed is low-cost, and the methods are easy to understand. Soft biometrics alone are not suitable for security related applications, but they can be used for improving the performance of traditional biometrics. A potentially feasible use for soft biometrics may be found in non-security, convenience type cases, such as domestic applications.

Increasing Security of Mobile Devices by Decreasing User Effort in Verification

Reliable user verification is important for security of computers and personal devices; however, most of well-performing verification methods require explicit user effort. As a consequence, an access is granted for a long time after the only successful verification, which allows replacing the authorized user to the advantage of an impostor, as it is often the case with mobile phones. This work proposes a method of frequent user verification, based on cascading of unobtrusive biometrics with more reliable biometrics, provided explicitly, in such a way that explicit effort is required only if unobtrusive verification fails. Experiments with voice, gait and fingerprint data have shown that in most of noise conditions cascade was able to satisfy security requirements of False Accept Rate 1% and to achieve overall False Reject Rate 3% or less, while requiring explicit effort in 10 - 60% of cases.

An affective model of user experience for interactive art

The development of Affective Interface technologies makes it possible to envision a new generation of Digital Arts and Entertainment applications, in which interaction will be based directly on the analysis of user experience. In this paper, we describe an approach to the development of Multimodal Affective Interfaces that supports real-time analysis of user experience as part of an Augmented Reality Art installation. The system relies on a PAD dimensional model of emotion to support the fusion of affective modalities, each input modality being represented as a PAD vector. A further advantage of the PAD model is that it can support a representation of affective responses that relate to aesthetic impressions.

Unobtrusive user identification with light biometrics

Biometric methods are used for recognition and verification of the identity of a person in many applications. Certain concerns over the obtrusive nature of their use, threats to privacy and even the danger of identity theft are rising. In this paper unobtrusive and privacy preserving light biometrics, such as height, weight, and body fat percentage are suggested for user identification. An experiment with 62 test subjects was conducted. In verification type of application total error rate of 11% was achieved using weight data alone and fusion with height data reduced the error rate to 2.4%. With a short list of five best scoring identities the percentage of cases with the correct identity on the list was 90% for weight alone and 100% for the combination of weight and height. The application domain for light biometrics is seen in non-security applications, such as homes, small offices and health clubs.

Personal video retrieval and browsing for mobile users

The latest camera-equipped mobile phones and faster cellular networks have increased the interest in mobile multimedia services. But for content consumption, delivery and creation, the limited capabilities of mobile terminals require special attention. This paper introduces the Candela platform, an infrastructure that allows the creation, storage and retrieval of home videos with special consideration of mobile terminals. Candela features a J2ME-based video recording and annotation tool which permits the creation and annotation of home videos on mobile phones. It offers an MPEG-7-based home video database which can be queried in an intelligent and user-oriented manner exploiting users’ personal domain ontologies. The platform employs terminal profiling techniques to deliver video retrieval user interfaces that personalize the search results according to the users preferences and terminal capabilities, facilitating effective retrieval of home videos via various both mobile and fixed terminals. For video playout, Candela features a meta player, a video player augmented by an interactive metadata display which can be used for fast content-based in-video browsing, helping to avoid the consumption and streaming of uninteresting video parts, thus reducing network load. Thereby, Candela forms a comprehensive video management platform for mobile phones fully covering mobile home video management from acquisition to delivery.

Unsupervised Speaker Change Detection for Mobile Device Recorded Speech

In this paper we propose an unsupervised speaker change detection (SCD) system developed for mobile device applications. We use Bayesian information criterion (BIC) to find initial speaker changes, which are then verified or discarded in the second phase by utilizing modified BIC and silence detector information. Silence information usage after initial BIC in decision making is useful to separate real changes from noise peaks. Enhanced peak detector adjusts BIC penalty parameter automatically, which improve the robustness and feasibility. Improved BIC based false alarm compensation (FAC) merges effectively consecutive segments belonging to same speaker. Our experiments have shown the robustness of the algorithm and it produces very satisfactory results for difficult mobile phone recorded speech data.

Mobile Video Capture Targeted Narrowband Audio Content Classification

Audio content analysis and automatic content management research field is facing new challenges from personal home video material created with mobile camera phones. We developed an audio content analysis and segmentation system that is robust to low sampling rate used in mobile phone camcorder tools and operates also well for AMR compressed data. Audio signal is segmented in five different classes, which can be used to classify videos for mobile video content management applications. The classification was done with Bayesian networks using topology of four-stage binary tree network that works in a hierarchical manner. Results were further smoothed within three second window using weighted sum of Bayesian networks class probabilities. The results show an average recognition accuracy of 88.6% for the high quality audio, 88.6% for the narrowband audio and accuracy of 82.9% for an AMR compressed audio. The effect of sampling rate to the recognition results is not significant and the effect of the AMR compression is also relatively low