Robust Gaze Estimation Based on Adaptive Fusion of Multiple Cameras

Gaze movements play a crucial role in human-computer interaction (HCI) applications. Recently, gaze tracking systems with a wide variety of applications have attracted much interest by the industry as well as the scientific community. The state-of-the-art gaze trackers are mostly non-intrusive and report high estimation accuracies. However, they require complex setups such as camera and geometric calibration in addition to subject-specific calibration. In this paper, we introduce a multi-camera gaze estimation system which requires less effort for the users in terms of the system setup and calibration. The system is based on an adaptive fusion of multiple independent camera systems in which the gaze estimation relies on simple cross-ratio (CR) geometry. Experimental results conducted on real data show that the proposed system achieves a significant accuracy improvement, by around 25%, over the traditional CR-based single camera systems through the novel adaptive multi-camera fusion scheme. The real-time system achieves less than 0.9 degrees accuracy error with very few calibration data (5 points) under natural head movements, which is competitive with more complex systems. Hence, the proposed system enables fast and user-friendly gaze tracking with minimum user effort without sacrificing too much accuracy.

Towards Convenient Calibration for Cross-Ratio based Gaze Estimation

Nuri Murat Arar, Hua Gao, Jean-Philippe Thiran

Eye gaze movements are considered as a salient modality for human computer interaction applications. Recently, cross-ratio (CR) based eye tracking methods have attracted increasing interest because they provide remote gaze estimation using a single uncalibrated camera. However, due to the simplification assumptions in CR-based methods, their performance is lower than the model-based approaches [8]. Several efforts have been made to improve the accuracy by compensating for the assumptions with subject- specific calibration. This paper presents a CR-based automatic gaze estimation system that accurately works under natural head movements. A subject-specific calibration method based on regularized least-squares regression (LSR) is introduced for achieving higher accuracy compared to other state-of-the-art calibration methods. Experimental results also show that the proposed calibration method generalizes better when fewer calibration points are used. This enables user friendly applications with minimum calibration effort without sacrificing too much accuracy. In addition, we adaptively fuse the estimation of the point of regard (PoR) from both eyes based on the visibility of eye features. The adaptive fusion scheme reduces accuracy error by around 20% and also increases the estimation coverage under natural head movements.

2015

Application of 3D range camera in virtual human-computer interfaces

Huan Du

In this work, we propose new ways to employ 3D-ranging systems in advanced human-computer interfaces and show that they can be used for precise hand-tracking systems aiming at virtual keyboard or mouse applications. We first implement a Structured Light (SL) based 3D-ranging system as the testbed for acquiring depth information. The ranging system uses an off-the-shelf projector combined with an industrial color camera, where the projector projects well-designed line patterns onto the scene and the camera captures the reflected light from the scene and the target. The depth information is recovered from the deformed line pattern and the pre-calibrated projector-camera system. With this system we can capture dense depth maps with moderate resolution and high frame rate as the simulated input to the hand-tracking system. We then explore an approach to 3D hand matching that combines a detailed 3D hand model (skeletal + polygonal model) with a Physical-based Model Fitting technique (PMF). Virtual forces are assigned between the depth measurements and the polygonal vertices of the model. The matching problem is formulated as the minimization of the distances between the measurement and the model surface with the guidance of virtual forces. This can employ 3D information simply and directly onto the parameter space, and yield good results under challenging conditions. To demonstrate the effectiveness of the approach we show that we can track hand motions such as typing or grasping frame by frame. The major limitation of this method is that it requires a good initialization of the hand model and the computation time to achieve convergence with good fitting results is quite long. We therefore investigated more complex hand models and feature-based matching techniques. A three-level hand model (skeletal + primitive + polygonal model) is proposed for the Multi-Level Feature Matching approach (MLFM). The matching is carried out from the bottom to top levels of different hand presentations, and distinctive features for each level extracted from both the measurements and the model are matched with Scaled Conjugate Gradient method (SCG). The matching result from lower level is used as the initial state of the model parameters for higher level matching. This hierarchical structure speeds up the optimization process and enhances the precision of the matching result. We demonstrate our approach with the same hand motion as in the previous method and compare the performance of these two methods. We show that the MLFM method is faster and more robust to hand occlusions than the PMF method. Finally, we present a more powerful approach based on the Back-Constrained Gaussian Process Latent Variable Model (BC-GPLVM) that import prior information of categorized hand motion such as typing. BC-GPLVM provides a low-dimensional embedding of hand motion data, with a density function that gives higher probability to gestures and motions close to normal typing motion. This model is learned from a small set of clusters of vast captured motion data. The prior model is then combined with the previous multi-level models and features to generate a uniform cost function for optimization. We show the tracking result to demonstrate that this method can be used for human-computer interfaces requiring precise 3D hand tracking.

EPFL2008

3D Gaze Estimation from Remote RGB-D Sensors

Kenneth Alberto Funes Mora

The development of systems able to retrieve and characterise the state of humans is important for many applications and fields of study. In particular, as a display of attention and interest, gaze is a fundamental cue in understanding people activities, behaviors, intentions, state of mind and personality. Moreover, gaze plays a major role in the communication process, like for showing attention to the speaker, indicating who is addressed or averting gaze to keep the floor. Therefore, many applications within the fields of human-human, human-robot and human-computer interaction could benefit from gaze sensing. However, despite significant advances during more than three decades of research, current gaze estimation technologies can not address the conditions often required within these fields, such as remote sensing, unconstrained user movements and minimum user calibration. Furthermore, to reduce cost, it is preferable to rely on consumer sensors, but this usually leads to low resolution and low contrast images that current techniques can hardly cope with. In this thesis we investigate the problem of automatic gaze estimation under head pose variations, low resolution sensing and different levels of user calibration, including the uncalibrated case. We propose to build a non-intrusive gaze estimation system based on remote consumer RGB-D sensors. In this context, we propose algorithmic solutions which overcome many of the limitations of previous systems. We thus address the main aspects of this problem: 3D head pose tracking, 3D gaze estimation, and gaze based application modeling. First, we develop an accurate model-based 3D head pose tracking system which adapts to the participant without requiring explicit actions. Second, to achieve a head pose invariant gaze estimation, we propose a method to correct the eye image appearance variations due to head pose. We then investigate on two different methodologies to infer the 3D gaze direction. The first one builds upon machine learning regression techniques. In this context, we propose strategies to improve their generalization, in particular, to handle different people. The second methodology is a new paradigm we propose and call geometric generative gaze estimation. This novel approach combines the benefits of geometric eye modeling (normally restricted to high resolution images due to the difficulty of feature extraction) with a stochastic segmentation process (adapted to low-resolution) within a Bayesian model allowing the decoupling of user specific geometry and session specific appearance parameters, along with the introduction of priors, which are appropriate for adaptation relying on small amounts of data. The aforementioned gaze estimation methods are validated through extensive experiments in a comprehensive database which we collected and made publicly available. Finally, we study the problem of automatic gaze coding in natural dyadic and group human interactions. The system builds upon the thesis contributions to handle unconstrained head movements and the lack of user calibration. It further exploits the 3D tracking of participants and their gaze to conduct a 3D geometric analysis within a multi-camera setup. Experiments on real and natural interactions demonstrate the system is highly accuracy. Overall, the methods developed in this dissertation are suitable for many applications, involving large diversity in terms of setup configuration, user calibration and mobility.