Damian Pęszor

Evaluation of Improvement in Orientation Estimation Through the Use of the Linear Acceleration Estimation in the Body Model

The need for broadly defined measures of human motion and estimation of motion parameters occurs in many research disciplines. The article concerns the evaluation of improvement in basic orientation estimation methods for IMU sensors, throughout the use of rigid body (segment in skeleton model) constraints. The verified method utilizes the correlation between rational motions and linear accelerations in the body model and reduces the impact of the external acceleration on the estimation of orientation. The experimental study concerns comparison of this method to other methods of leveling the influence of linear external acceleration.

Synchronization of motion sequences from different sources

The paper describes an algorithm for the synchronization of motion sequences derived with different motion capture (mocap) systems. The algorithm is invented for the temporal matching of the motion represented as angular orientation timeseries obtained with different mocap systems. We employed the PCA to reduce problem to a single dimension, then the algorithm comprises twofold exhaustive search allowing for the precise matching of sequences. The method was verified with both - semi synthetic and real sequences.

Automatic Markers’ Influence Calculation for Facial Animation Based on Performance Capture

Marker-based performance capture is a technique that enables acquisition of expression and mimicry of human face. This data can be used to propel facial animation system, be it bone driven or similarly dependant on position of points in space. Every model that is to be animated has to be analyzed in order to select level of influence each marker has over each vertex of said model. This process can be quite tedious if done manually. In this paper we present an approach for automatic calculation of markers’ influence based on position of vertex on human face’s surface obtained by acquisition using structured light-based scanner or similar approach.

Estimation of marker placement based on fiducial points for automatic facial animation

Nowadays, facial animation is achieved using performance capture techniques, of which many is based on the concept of tracking reflective markers placed on human face. Currently available systems either recognize actions which are used to propel final animation or provide information about artificial skeleton which animator is using to animate character. Creation of fully automated system for facial animation requires a way to tie surface on which marker is placed directly to surface of animated model representing different face, so actor’s facial expressions could be applied to animated model without need for adjustment. This paper presents a way of estimating marker placement on realistic model of human face obtained using structured light-based 3D scanner. Preprocessed mesh is consequently aligned to neutral model using Trimmed and classic version of Iterative Closest Point algorithm, while fiducial points are found using Anthroface 3D or similar methods related toface recognition until finally marker ...

Preprocessing of 3D scanned images for facial animation on the basis of realistic acquisition

Facial animation done by performance capture techniques is a subject of growing importance while the quality of graphics and animation in entertainment industry is constantly improving. Most realistic results in facial animation can be obtained using realistic acquisition of human face by various three-dimensional scanning devices. Such registration, however, is prone to errors related to specifity of scanned area, therefore some preprocessing is needed so the obtained model could be used. The aim of this paper is to present typical, face-specific issues as well as solutions related to preprocessing of mesh constructed by scanning techniques. Mesh traversing is applied to reduce number of noisy data related to light dispersion and reflection. Non-manifold edges and vertices are corrected on basis of specifity of studied area, strips of noisy triangles typical for hair are removed and holes typical for chin-neck part of model are filled. Resulting mesh represents single, continuous surface without non-mani...

Reference Data Set for Accuracy Evaluation of Orientation Estimation Algorithms for Inertial Motion Capture Systems

In this paper, we describe the freely available repository, RepoIMU, proposed for the needs of orientation estimation algorithm evaluation for inertial motion capture systems. Such algorithms fuse data from IMU sensors - accelerometer, magnetometer and gyroscope. Although such systems gain in popularity, the evaluation of methods used in their development is still not well covered in literature. In presented repository, unlike many others, signals from IMU sensors combined into easily interpretable kinematic chain are synchronized, and not only cropped to, reference orientation obtained from accurate optical motion capture system.

Precision of Gait Indices Approximation by Kinect Based Motion Acquisition

Step length constitutes one of the important gait indices. The research described in the present paper was focused on the method of determination of the step length by means of the Kinect device. Gait sequences recorded in the Human Motion Laboratory of the Polish-Japanese Academy of Information Technology using the Vicon system played the role of the reference data. The group of six subjects participated in the experiments. Conclusions from the comparative analysis of the results of both approaches summarize the paper.

Fast Moving UAV Collision Avoidance Using Optical Flow and Stereovision

Unmanned aerial vehicles are becoming popular, but their autonomous operation is constrained by their collision avoidance ability in high-velocity movement. We propose a simple collision avoidance scheme for fast, business-grade fixed-wing aircraft which is based on optical flow and stereovision. We calculate optical flow on parts of the image that are essential for collision avoidance and enlarge the analysed area only as long, as the framerate allows, thus avoiding the need to stretch calculations over several frames.

Zero-Velocity Detectors for Orientation Estimation Problem

The paper concerns the application of the zero-velocity updates in orientation estimation filter for a inertial motion capture system. The evaluation of three commonly used detectors together with quaternion complementary filter to orientation estimation based on indications of gyroscope, magnetometer and accelerometer sensors is presented.

Facial Reconstruction on the Basis of Video Surveillance System for the Purpose of Suspect Identification

Growing importance and commonness of video surveillance systems brings new possibilities in the area of crime suspect identification. While suspects can be recognized on video recordings, it is often a difficult task, because in most cases parts of suspect’s face are occluded. Even if there are multiple cameras, and the recordings are long enough to expose entirety of suspect’s face, it is challenging for an observer to accumulate information from different cameras and frames. We propose to solve this problem by reconstructing a three-dimensional mesh that could be presented to an observer, so he could identify suspect based on accumulated information rather than fragmented one, while choosing any angle of observation. Our approach is based on extraction of anthropological features, so that even with imperfect recordings, the most important features in terms of facial recognition are preserved, while those not registered might be supplemented with generic facial surface.