AbstractSmartphone applications (SPA) now offer the ability to provide accessible in-home monitoring of relevant individual health biomarkers. Previous cross-sectional validations of similar technologies have reported acceptable accuracy with high-grade body composition assessments; this research assessed longitudinal agreement of a novel SPA across a self-managed weight loss intervention of thirty-eight participants (twenty-one males, seventeen females). Estimations of body mass (BM), body fat percentage (BF%), fat-free mass (FFM) and waist circumference (WC) from the SPA were compared with ground truth (GT) measures from a dual-energy X-ray absorptiometry scanner and expert technician measurement. Small mean differences (MD) and standard error of estimate (SEE) were observed between method deltas (ΔBM: MD = 0·12 kg, SEE = 2·82 kg; ΔBF%: MD = 0·06 %, SEE = 1·65 %; ΔFFM: MD = 0·17 kg, SEE = 1·65 kg; ΔWC: MD = 1·16 cm, SEE = 2·52 cm). Concordance correlation coefficient (CCC) assessed longitudinal agreement between the SPA and GT methods, with moderate concordance (CCC: 0·55–0·73) observed for all measures. The novel SPA may not be interchangeable with high-accuracy medical scanning methods yet offers significant benefits in cost, accessibility and user comfort, in conjunction with the ability to monitor body shape and composition estimates over time.

Background/objectives: Body composition and anthropometry assessment from two-dimensional smartphone images is possible through advancement of computational hardware and artificial intelligence (AI) techniques. This study established agreement of a novel smartphone assessment, compared with traditional bioelectrical impedance analysis (BIA), and criterion measures. Subjects/methods: Body composition of 929 adults was measured using DXA (GE lunar iDXA), a foot-to-foot BIA machine (TANITA BC-313), and predictions from two-dimensional smartphone images. Anthropometry measures were also collected. Body composition and anthropometry estimates were compared via concordance coefficient correlation (CCC), equivalence testing, Bland-Altman analysis, and root mean square error (RMSE). Results: 2D smartphone image predictions for percent body fat (%BF) (males: CCC = 0.90 and RMSE = 2.9, and females: CCC = 0.90 and RMSE = 2.8) reported greater agreement with DXA measures than the BIA measures (males: CCC = 0.66 and RMSE = 5.6, and females: CCC = 0.79 and RMSE = 4.6). All anthropometry 2D smartphone image predictions had a strong agreement with criterion measurements (CCC >= 0.84 and RMSE <= 3.3). Body composition and anthropometry measures predicted by the 2D smartphone images were clinically equivalent at +/- 2.5 and +/- 5.0% thresholds. BIA %BF was not equivalent at either threshold; with only female BIA fat-free mass equivalent at the +/- 5% threshold. Conclusion: Body composition predictions from 2D smartphone application images provide a promising alternative to BIA scales for in-home body composition assessment. Future research should assess the validity of this method for longitudinally tracking body composition and indicating an individual's potential risk of chronic diseases.

Limited reference values exist for visceral adipose tissue (VAT) mass measured by DXA. The objectives of this study were to provide reference values for DXA-derived VAT mass and compare the association with anthropometry measures. The study cohort comprised 677 men and 738 women aged 18–65 years from Western Australia. Whole-body scans using a GE Lunar iDXA and anthropometry measures were collected. Reference percentile data were stratified by sex and age. Correlation analysis compared DXA-derived and anthropometry variables. Specificity, sensitivity, and Youden’s Index were used to evaluate the ability of anthropometric thresholds to predict individuals with high VAT. In men, waist circumference (WC), waist-hip ratio, and waist-height ratio (WHtR) had ‘high’ correlations with VAT mass. In women, only WHtR was ‘highly’ correlated with VAT mass. Overweight thresholds for WC, along with a body mass index of 25.0 kg/m 2 in women, had the highest combination of sensitivity and specificity when using anthropometry measures to identify individuals with high VAT mass. We provide the first reference data sets for DXA-derived VAT mass among Western Australians. Excessive VAT mass may be identified in men using the overweight WC threshold and in women using both the overweight BMI and WC thresholds.

This paper presents a novel personal identification and verification system using information extracted from the hand shape and texture. The system has two major constituent modules: a fully automatic and robust peg free segmentation and pose normalisation module, and a recognition module. In the first module, the hand is segmented from its background using a thresholding technique based on Otsu`s method combined with a skin colour detector. A set of fully automatic algorithms are then proposed to segment the palm and fingers. In these algorithms, the skeleton and the contour of the hand and fingers are estimated and used to determine the global pose of the hand and the pose of each individual finger. Finally the palm and fingers are cropped, pose corrected and normalised. In the recognition module, various shape and texture based features are extracted and used for matching purposes. The modified Hausdorff distance, the Iterative Closest Point (ICP) and Independent Component Analysis (ICA) algorithms are used for shape and texture features of the fingers. For the palmprints, we use the Discrete Cosine Transform (DCT), directional line features and ICA. Recognition (identification and verification) tests were performed using fusion strategies based on the similarity scores of the fingers and the palm. Experimental results show that the proposed system exhibits a superior performance over existing systems with an accuracy of over 98\% for hand identification and verification (at equal error rate) in a database of 560 different subjects.

INTRODUCTION Forward dynamic simulations run with musculoskeletal models in OpenSim can shed some light on the multiple, and often non-intuitive, muscle activation strategies employed to support ACL loading during non-contact sidestepping in sport. However, the validity of the musculoskeletal model for forward dynamic simulations may be compromised if the adopted segmental inertial parameters (IP) are not participant-specific. More specifically, inaccurate segmental IP may lead to inconsistencies between the measured ground reaction forces (GRF) and the model kinematics used to find a pattern of muscle excitations via the optimization solution. By default, OpenSim re-scales a generic musculoskeletal model to predict the IP of the participant. Although segmental IP of elite athletes are shown to be not accurately predicted using generic scaling equations [1], the effect of inaccurate segmental IP on the validity of the musculoskeletal model remains unknown.

This paper examines the accuracy of body segment inertia tensors estimated by combining information from dual-energy X-ray absorptiometry and a three-dimensional modelling technique proposed by Zatsiorsky et al. (1 990) (DXAIVol method). The inertia tensor of a frozen pig cadaver was estimated using the novel DXAlVol method and traditional compound pendulum techniques. The pig cadaver was projected through the air and the experimental 'ground truth kinematics' were recorded. Simulated kinematics of the pig cadaver flight were generated using the inertia tensor derived from the DXAlVol and compound pendulum methods and compared to the ground truth kinematics. Simulations based on the novel D W o l method's inertia tensor traded the experimentally recorded flight of the frozen pig cadaver with superior accuracy.

The aims of this study were to: (i) establish a new criterion method to validate inertia tensor estimates by setting the experimental angular velocity data of an airborne objects as ground truth against simulations run with the estimated tensors, and (ii) test the sensitivity of the simulations to changes in the inertia tensor components. A rigid steel cylinder was covered with reflective kinematic markers and projected through a calibrated motion capture volume. Simulations of the airborne motion were run with two models, using inertia tensor estimated with geometric formula or the compound pendulum technique. The deviation angles between experimental (ground truth) and simulated angular velocity vectors and the root mean squared deviation angle were computed for every simulation. Monte Carlo analyses were performed to assess the sensitivity of simulations to changes in magnitude of principal moments of inertia within ±10% and to changes in orientation of principal axes of inertia within ±10° (of the geometric-based inertia tensor). Root mean squared deviation angles ranged between 2.9° and 4.3° for the inertia tensor estimated geometrically, and between 11.7° and 15.2° for the compound pendulum values. Errors up to 10% in magnitude of principal moments of inertia yielded root mean squared deviation angles ranging between 3.2° and 6.6°, and between 5.5° and 7.9° when lumped with errors of 10° in principal axes of inertia orientation. The proposed technique can effectively validate inertia tensors from novel estimation methods of body segment inertial parameter. Principal axes of inertia orientation should not be neglected when modelling human/animal mechanics.

The paper proposes a method for robust face recognition from low quality Kinect acquired images which have a wide range of variations in head pose, illumination, facial expressions, sunglass disguise and occlusions by hand. Multiple Kinect images of a person are considered as an image set and face recognition from these images is formulated as an RGB–D image set classification problem. The Kinect acquired raw depth data is used for pose estimation and an automatic cropping of the face region. Based upon the estimated poses, the face images of a set are divided into multiple image subsets. An efficient block based covariance matrix representation is proposed to model images in an image subset on Riemannian manifold (Lie group). For classification, SVM models are separately learnt for each image subset on the Lie group of Riemannian manifold and a fusion strategy is introduced to combine results from all image subsets. The proposed technique has been evaluated on a combination of three large data sets containing over 35,000 RGB–D images under challenging conditions. The proposed RGB–D based image set classification incurs low computational cost and achieves an identification rate as high as 99.5%.

In this paper, we present a fully automated multimodal Curvelet-based approach for textured 3D face recognition. The proposed approach relies on a novel multimodal keypoint detector capable of repeatably identifying keypoints on textured 3D face surfaces. Unique local surface descriptors are then constructed around each detected keypoint by integrating Curvelet elements of different orientations, resulting in highly descriptive rotation invariant features. Unlike previously reported Curvelet-based face recognition algorithms which extract global features from textured faces only, our algorithm extracts both texture and 3D local features. In addition, this is achieved across a number of frequency bands to achieve robust and accurate recognition under varying illumination conditions and facial expressions. The proposed algorithm was evaluated using three well-known and challenging datasets, namely FRGC v2, BU-3DFE and Bosphorus datasets. Reported results show superior performance compared to prior art, with 99.2%, 95.1% and 91% verification rates at 0.001 FAR for FRGC v2, BU-3DFE and Bosphorus datasets, respectively.

We have made a research on the field of biometrics and found that multimodal biometric systems plays a major role in authentication and security for different types of areas such as science, medicine, engineering, robotics, etc. Face recognition using biometrics is the frontier of the security systems. The ear, which is a biometric identifier, that has some desirable properties such as universality, uniqueness, permanence, can also be used for security purposes. We propose a multi-modal biometric system combining face (2D+3D) and ear features at various levels which uses Microsoft Kinect. Commencing with the survey of existing algorithms, applied to face (2D+3D) and ear data, we focused on fast discrete curvelet transformation techniques for face (2D+3D) recognition and active contour algorithm techniques for ear recognition. Finding optimal fusion level and avoiding redundancy in the extracted features are some challenges in designing our system. Our system is insensitive to lighting conditions, pose variations, aging and can completely replace the current recognition systems economically and provide a better security. A total of 250 subjects participated in data acquisition sessions through kinect. The results are obtained separately each biometric and fused at metric level for higher accuracy. Our multimodal algorithm performed better by achieving 97 percent and 95 percent verification rates with 0.01 false acceptance rate which is greater than either 2D face or 3D face or ear recognition algorithm alone in a statistical and significant manner. Keywords—Biometrics; Security System; Person Identification; Modalities; Kinect; Feature Extraction; Curvelets; Eigen Faces; Geometric Features; Feature level Fusion.

In this letter, we present a robust single modality feature-based algorithm for 3-D face recognition. The proposed algorithm exploits Curvelet transform not only to detect salient points on the face but also to build multi-scale local surface descriptors that can capture highly distinctive rotation/displacement invariant local features around the detected keypoints. This approach is shown to provide robust and accurate recognition under varying illumination conditions and facial expressions. Using the well-known and challenging FRGC v2 dataset, we report a superior performance compared to other algorithms, with a 97.83% verification rate for probes with all facial expressions.

We investigate the ability of Curvelet transform to extract robust features from 3D faces under different facial expressions and pose variations. We also investigate the contribution of fusing 3D and 2D data to achieve multimodal face identification using Curvelet features to describe semi-rigid regions. Extracted Curvelet features are filtered before applying PCA for dimensionality reduction. In the testing phase, a similarity matrix is built by matching faces from probe and gallery. Four different matching processes (with respect to the type of data and the regions of interest) were considered. The resulting similarity matrices were fused at the score level to generate the proposed Curvelet-based multimodal face identification system. Our system was evaluated using the well-known and challenging FRGC v2 dataset. Reported experimental results show that the proposed Curvelet-based approach provides superior performance for non-neutral facial expressions, with an identification rate of 93.74%. For neutral expressions, the system is shown to exhibit an identification rate of 97.43%.

This paper presents a fully automatic system which exploits the dynamics of 3D videos and is capable of recognizing six basic facial expressions. Local video-patches of variable lengths are extracted from different locations of the training videos and represented as points on the Grass-mannian manifold. An efficient spectral clustering based algorithm is used to separately cluster points for each of the six expression classes. The resulting cluster centers are matched with the points of a test video and a voting based strategy is used to decide about the expression class of the test video. The proposed system is tested on the largest publicly available 3D video database, BU4DFE. The experimental results show that the system achieves a very high classification accuracy and outperforms the current state of the art algorithms for facial expression recognition from 3D videos.

As accurate body segment inertial parameters (BSIPs) are difficult to obtain in motion analysis, this study computed individual BSIPs from DXA scan images. Therefore, by co-registering areal density data with DXA grayscale image, the relationship between pixel color gradient and the mass within the pixel area could be established. Thus, one can calculate BSIPs, including segment mass, center of mass (COM) and moment of inertia about the sagittal axis (Ixx). This technique calculated whole body mass very accurately (%RMSE of < 1.5%) relatively to results of the generic DXA scanner software. The BSIPs of elite male and female swimmers, and young adult Caucasian males (n = 28), were computed using this DXA method and 5 other common indirect estimation methods. A 3D surface scan of each subject enabled mapping of key anthropometric variables required for the 5 indirect estimation methods. Mass, COM and Ixx were calculated for seven body segments (head, trunk, head + trunk, upper arm, forearm, thigh and shank). Between-group comparisons of BSIPs revealed that elite female swimmers had the lowest segment masses of the three groups (p < 0.05). Elite male swimmers recorded the greatest inertial parameters of the trunk and upper arms (p < 0.05). Using the DXA method as the criterion, the five indirect methods produced errors greater than 10% for at least one BSIP in all three populations. Therefore, caution is required when computing BSIPs for elite swimmers via these indirect methods, DXA accurately estimated BSIPs in the frontal plane. Key PointsElite swimmers have significantly different body segment inertial parameters than young adult Caucasian males.The errors computed from indirect BSIP estimation methods are large regardless whether applied to elite swimmers or young adult Caucasian males.No indirect estimation method consistently performed best.

Local surface description is a critical stage for feature matching and recognition tasks. This paper presents a rotation invariant local surface descriptor, called 3D-Div. The proposed descriptor is based on the concept of 3D vector field's divergence, extensively used in electromagnetic theory. To generate a 3D-Div descriptor of a 3D surface, a local surface patch is parameterized around a randomly selected 3D point at a fixed scale. A unique Local Reference Frame (LRF) is then constructed at that 3D point using all the neighboring points forming the patch. A normalized 3D vector field is then computed at each point in the patch and referenced with LRF vectors. The 3D-Div descriptor is finally generated as the divergence of the reoriented 3D vector field. We tested our proposed descriptor on the challenging low resolution Washington RGB-D (Kinect) object dataset, for the task of automatic 3D object recognition. Reported experimental results show that 3D-Div based recognition achieves 93% accuracy as compared to 85% for existing state-of-the-art depth kernel descriptors [2].

In this paper, we present a method to combine a Gaussian Process regression and a particle filter to track the 3D human pose in video sequences. We first build the probabilistic discriminative model that maps the silhouette descriptor to multiple 3D human poses using a Gaussian Process regression. The multimodal output distribution from the Gaussian Process regression are combined with the particle filter to track the 3D human pose in each frame of the video sequence. The predictions from the discriminative model are used to generate the hypothesis space for the particle filter and to initialize the tracking. We evaluate our approach on the HumanEva-I dataset and on the video sequences of Parkinson's patients. The evaluation results show that our approach does not require initialization and successfully tracks the 3D human pose over long video sequences.

Object detection is an important step towards object recognition. A robust object detection system is one that can detect an object of any class. In this paper, we present a fully automatic approach to object detection based on an objectness measure. The proposed automatic object detection approach quantifies the likelihood for an image window to encompass objects in the image. It can discriminate between multiple objects in a scene, with individual windows capturing each detected object. Most importantly, the proposed approach does not require any manual input. We tested this approach on the challenging PASCAL VOC 07 dataset. Experimental results show that our approach provides a more accurate estimation of the required number of windows for an input image. The proposed technique is computationally efficient and takes less than 4 sec. per image.