Project topics and materials
Get Complete Project Material File(s) Now! »
Body soft biometrics
Height, gait, body weight and color of clothes concern the body and are the main traits that can be extracted from a distance. The best distinctiveness is provided by gait detection, which is why gait is occasionally referred to as a classical biometric.
Anthropometric measures
The studies on anthropometric measures are not generally driven by biometric use. While at the beginning anthropometry was a technique used in physical anthropology to study the physical development of the human species; nowadays it is employed in industrial/clothing design, ergonomics, and architecture to optimize the products to the customers needs. Other interesting works regards the study of population statistics so as to monitor changes in lifestyle, and nutrition to track body dimensions (e.g. obesity epidemic) [21].
The first biometric application of anthropometry is due to Alphonse Bertillon. His anthropometry-based classification method was used to identify criminals, and it is indeed one of the few examples of anthropometric measure used as biometric identifier.
After the historical contribution of Bertillon, one of the first works that tried to estimate anthropometric measures from images is the one presented in [22]. The authors of this paper, use a priori statistical information about the human body, to establish the correspondence between a set of manually marked points and the segments that compose the body parts. In a second step a set of postures is considered and finally pose and anthropometric measurements are obtained.
Results are achieved by minimizing an appropriate cost function and according to a model inspired by human body statistics collected for medical research. The recent commercialization of millimeter wave scanners and full 3D body scanners has raised the interest of the research community. Some works have suggested that the idea of anthropometry based people identification is feasible and proposed some approaches. An example is the work of [23] where the authors investigate the utility of 1D anthropometric measurements as a biometric trait for human identification. They analyze 27 measurements from 2144 subjects, by reducing those measures to a smaller set thanks to dimensionality reduction techniques they obtain rank-1 identification of 83% and 94% using just ten and fifteen features.
Other interesting works on anthropometric measures are shown in [24] where height, stride, and other measures are taken into account for people identification; and in [25] where anthropometric measures are estimated from calibrated monocular sequences. By tracking subjects across multiple cameras the authors estimate stature, shoulder breadth, and link them with specific features provided by gait to perform people identification.
Height
Even if height is part of the more general anthropometric measures, we dedicate to it a separate section because the computer vision community explored deeply its extraction and possible applications.
Height estimation is an already mature topic in the literature and it has been exploited several times. One of the earliest approaches is presented in [16], the authors use the content of the image to compute geometrical properties of objects that lie on the same plane, later they can compare objects dimensions. By knowing the height of given objects in the scene they are able to measure height of people in the camera field of view (FOV). Extending this last work, the authors of [26] propose further improvements using multiple measures and a statistical approach to remove outliers, using the proposed approach they arrive to a precision of 1cm for subjects walking in an unconstrained scenario. Precise measurement of height has been already used in combination with
other features so as to track people across multiple camera systems, and to allow the identification of the same person in multiple video streams [27]. The estimation is performed via the computation of height related to the real world coordinates estimated in camera images.
Height is possibly one of the most used in real cases and can become under certain circumstances a crime evidence. It is indeed one of the main factors used in photogrammetry. This technique is nowadays widely used to estimate anthropometric measures from images or video surveillance footages. The Netherlands Forensic Institute has performed a comparison [28] of two methods for obtaining body height measurements from images. One is based on projective geometry and the other one on 3D modeling of the crime scene. Keeping the same camera settings setup, the authors demonstrate that the predictions of both methods are accurate, but changing camera position makes the first algorithm less reliable. Moreover, also the 3D reconstruction of the environment can be helpful as this kind of analysis greatly simplify the extraction of measurements. The possibility of using such a technique is explored in [29] where the authors use landmarks within the scene to enable the automatic collection of subjects’ height measurements.
Weight
Since the beginning weight was introduced within the list of the soft biometric traits [12]. However it was not fully explored as soft biometric traits. A field where weight is considered an important feature is represented by medical studies. Here the main interest is represented by the visual extraction capability and reliability of human operators in case of emergency situations where there is no possibility of using scales like in [2, 30, 31].
Other interests are represented by the use of weight as a foremost important feature that helps to monitor body health status [32]. Additionally, a branch of medical studies explore the forensic aspect of weight estimation [33] so as to recover information from latent traces that help to recognize victims or crime suspects.
To the best of our knowledge, the only paper which involves weight directly referring to it as a soft biometrics is [34], where the authors use a scale to weigh clients of a fingerprint recognition system. By exploiting weight and body fat measurements the authors reduce the total error rate of the system by 2.4%. Another work [5] considered weight as a discrete value visually defined by subjects participating to a psycho-visual experiment. However, the values used (Very Thin, Thin, Average, Fat, Very Fat) show that rather than the weight itself, the description refers to the way fat is distributed on the inspected body. That is to say users described the body build of subjects rather than their body mass. A similar experiment is reported on [35] where the authors propose other features alongside with weight. Additionally, the work points out the relevance of this trait for eyewitness testimony. Some experiments are performed by [36, 37] that involve fitting a 3D human model to a point cloud obtained in the first case by a RGBD camera, and in the second case by a series of stereo cameras. In this case extracting the weight is straightforward if we consider the average density of the human body. While in the first work the weight (and gender) estimation is an interesting side effect, in the second case the authors purposely try to extract such information.
Table of contents :
1 Introduction
1.1 Motivations, objectives and contributions
1.2 Thesis organization
2 Soft Biometric: a state of the art
2.1 Introduction
2.2 New definition of Soft Biometrics
2.3 Traits
2.3.1 Body soft biometrics
2.3.1.1 Anthropometric measures
2.3.1.2 Height
2.3.1.3 Weight
2.3.1.4 Gender
2.3.1.5 Gait
2.3.2 Face soft biometrics
2.3.2.1 Color based
2.3.2.2 Gender
2.3.2.3 Beard and Mustache detection
2.3.2.4 Age
2.3.2.5 Ethnicity
2.3.2.6 Facial measurements
2.3.3 Accessory soft biometrics
2.3.3.1 Clothes color and clothes classification
2.3.3.2 Eye Glasses detection
2.3.4 Traits instances and direct measures
2.4 Techniques and applications of soft biometrics
2.4.1 Pruning
2.4.2 Identification
2.4.3 Semantic description
3 Body soft biometrics extraction
3.1 Anthropometry and biometrics
3.2 Height estimation
3.3 Weight estimation
3.3.1 Related works
3.3.2 Inferring the model
3.3.3 Sources of data
3.3.4 Experimental results
3.3.4.1 Results under ideal conditions
3.3.4.2 Biased measures analysis
3.3.4.3 Real case analysis
3.4 Gender classification
3.5 Conclusion
3.5.1 Critical analysis of the contributions
4 Pruning search in databases
4.1 Introduction
4.2 Previous work
4.2.1 Anthropometry-based recognition
4.2.2 Soft-biometrics based database pruning
4.3 Proposed case study
4.4 Experimental results
4.4.1 Performance analysis of the anthropometry system
4.4.2 Recognition accuracy increase by pruning
4.5 Conclusion
4.5.1 Critical analysis of the contributions
5 Identification
5.1 Introduction
5.2 Linking video camera outputs and building map
5.2.1 Homography mapping
5.2.2 Single camera tracking
5.3 Soft biometrics based people re-identification
5.3.1 Height
5.3.2 Weight
5.3.3 Clothes color
5.3.4 Bag of body soft biometrics for re-identification
5.4 Displaying appropriate information
5.5 Conclusion
5.5.1 Critical analysis of the contributions
6 Semantic annotation and monitoring
6.1 Telemedicine
6.1.1 Introduction
6.1.2 Subjective and Objective Ideal weight
6.1.3 Weight estimation
6.1.4 Application description
6.1.5 Conclusion
6.2 Applications to space
6.2.1 Astronauts’ weight
6.2.2 Vision Based Weight Estimation
6.2.3 Further Work and Discussion
6.2.4 Critical analysis of the contributions
7 Conclusion
7.1 Future Works
7.1.1 New traits
7.1.2 Database
7.1.3 New sensors and technologies
7.1.4 Research studies
A French Version : Introduction
A.1 Motivations, objectifs et contributions
A.2 Organisation de la thèse
B Biométrie douce : état de l’art
B.1 Introduction
B.2 Nouvelle définition de la biométrie douces
B.3 Traits
B.3.1 Biométrie douce pour le corps
B.3.1.1 Les mesures anthropométriques
B.3.1.2 Taille
B.3.1.3 Poids
B.3.1.4 Genre
B.3.1.5 Démarche
B.3.2 Biométrie douce du visage
B.3.2.1 Couleur
B.3.2.2 Sexe
B.3.2.3 Détection de barbe et moustache
B.3.2.4 Age
B.3.2.5 Ethnie
B.3.2.6 Mesures du visage
B.3.3 Biométrie douce et accessoires
B.3.3.1 La couleur et la classification des vêtements
References
