Ivar Austvoll

Fuzzy inference-based reliable fall detection using kinect and accelerometer

Falls are major causes of mortality and morbidity in the elderly. However, prevalent methods only utilize accelerometers or both accelerometers and gyroscopes to separate falls from activities of daily living. This makes it not easy to distinguish real falls from fall-like activities. The existing CCD-camera based solutions require time for installation, camera calibration and are not generally cheap. In this paper we show how to achieve reliable fall detection. The detection is done by a fuzzy inference system using low-cost Kinect and a device consisting of an accelerometer and a gyroscope. The experimental results indicate high accuracy of the detection and effectiveness of the system.

A study of the yosemite sequence used as a test sequence for estimation of optical flow

Since the publication of the comparative study done by Barron et al. on optical flow estimation, a race was started to achieve more and more accurate and dense velocity fields. For comparison a few synthetic image sequences has been used. The most complex of these is the Yosemite Flying sequence that contains both a diverging field, occlusion and multiple motions at the horizon. About 10 years ago it was suggested to remove the sky region because the correct flow used in earlier work was not found to be the real ground truth for this region. In this paper we present a study of the sky region in this test sequence, and discuss its usefulness for evaluation of optical flow estimation.

Particle swarm optimization with soft search space partitioning for video-based markerless pose tracking

This paper proposes a new algorithm called soft partitioning particle swarm optimization (SPPSO), which performs video-based markerless human pose tracking by optimizing a fitness function in a 31-dimensional search space. The fitness function is based on foreground segmentation and edges. SPPSO divides the optimization into two stages that exploit the hierarchical structure of the model. The first stage only optimizes the most important parameters, whereas the second is a global optimization which also refines the estimates from the first stage. Experiments with the publicly available Lee walk dataset showed that SPPSO performs better than the annealed particle filter at a frame rate of 20 fps, and equally well at 60 fps. The better performance at the lower frame rate is attributed to the explicit exploitation of the hierarchical model structure.

Directional filters and a new structure for estimation of optical flow

We study the use of directional filters with application to estimation of dense velocity fields. We suggest a new type of directional filters based on the discrete prolate spheroidal sequence (DPSS). These filters are used for two purposes, decomposition of images in subimages with directional information, and a set of quadrature filters based on the DPSS is used for estimation of direction. For efficient computation of optical flow we suggest a new structure using 2D directional filters. After a set of scale space filters the image sequence is decomposed in subsequences with directional properties. For each direction phase space-time slices is extracted and a set of directional quadrature filters is used to estimate component velocities. The component velocities are finally used for estimation of a full velocity field. Compared to other methods our system is among the best, when accuracy, density and computational complexity is considered.

Region covariance matrix-based object tracking with occlusions handling

This work proposes an optical-flow based feature tracking that is combined with region covariance matrix for dealing with tracking of an object undergoing considerable occlusions. The object is tracked using a set of key-points. The key-points are tracked via a computationally inexpensive optical flow algorithm. If the occlusion of the feature is detected the algorithm calculates the covariance matrix inside a region, which is located at the features position just before the occlusion. The region covariance matrix is then used to detect the ending of the feature occlusion. This is achieved via comparing the covariance matrix based similarity measures in some window surrounding the occluded key-point. The outliers that arise in the optical flow at the boundary of the objects are excluded using RANSAC and affine transformation. Experimental results that were obtained on freely available image sequences show the feasibility of our approach to perform tracking of objects undergoing considerable occlusions. The resulting algorithm can cope with occlusions of faces as well as objects of similar colors and shapes.

Dense motion field estimation using spatial filtering and quasi eigenfunction approximations

Earlier research on a particular phase based method for estimation of dense motion fields show promising results for complicated sequences. Typical for phase based methods is a decomposition of the sequence into directional signals by directional filters. From each directional signal a component velocity can be estimated. In this context we propose a method for estimating component velocities which further improves the accuracy, by modelling 2D space-time slices in the direction of the filters as single component AM-FM modulated signals. A method for demodulation based on quasi eigenfunction approximation is the basis for estimation of component velocities, along with a tensor formulation and eigenvector analysis of this tensor. We demonstrate the method on the well known reference sequence Yosemite fly-by and the Rubics cube sequence.

Filter banks, wavelets, and frames with applications in computer vision and image processing (a review)

The purpose of this article is to present some of the fundamental principles of filter banks, wavelets and frames and their connections, with special emphasis on applications in computer vision and image processing. This is a vast field and we can only give a glimpse of it. We start with a short historical review and a rather broad discussion of filter banks, wavelets and frames. It is discussed how filter banks and wavelets are connected via multiresolution. Some of the most important structures and properties are presented but hardly no mathematical details are given. We focus especially on directional filter banks and wavelets, on analysis and extraction of directional features in images and image sequences. A system for motion estimation (estimation of optical flow) is presented.

Motion based detection of respiration rate in infants using video

Worldwide, 11% of infants are born prematurely, with a substantially increased risk of apneas, i.e. suspension of breathing, and in need of extra care after birth. Sensors used in neonatal intensive care units today are attached to the infant and can cause false alarms due to contact difficulties. In this paper we propose to monitor the respiratory rate of infants by using video recordings and video processing. A phase-based algorithm for the detection of small and repetitive motions is combined with a method for detecting situations where the camera view is blocked by a caretaker. The phase-based respiration detection (PRD) is tested on a small test set of videos of sleeping babies, where each breath is manually marked providing a truth signal. Compared to a difference-based method, PRD is much more robust and promising. The PRD is also tested on two adults, and the resulting respiratory rate is compared to a truth signal extracted from measuring the impedance pneumography of the thorax with encouraging results.

Detecting Chest Compression Depth Using a Smartphone Camera and Motion Segmentation

Telephone assisted guidance between dispatcher and bystander providing cardiopulmonary resuscitation (CPR) can improve the quality of the CPR provided to patients suffering from cardiac arrest. Our research group has earlier proposed a system for communication and feedback of the compression rate to the dispatcher through a smartphone application. In this paper we have investigated the possibilities of providing the dispatcher with more information by also detecting the compression depth. Our method involves detection of bystander‘s position in the image frame and detection of compression depth by generating Accumulative Difference Images (ADIs). The method shows promising results and give reason to further develop a general and robust solution to be embedded in the smartphone application.

Image Processing of Leaf Movements in Mimosa pudica

In this paper the plant Mimosa pudica’s response to changed illumination conditions is being examined. An image processing routine, using the HSV color model and triangle intensity threshold segmentation, is developed to segment time-lapse image series of Mimosa pudica, quantifying the plant’s image pixel count as a measure of movement. Furthermore, the method of Farneback is used to estimate dense optical flow (both magnitude and direction), describing the plants movement orientation in the image plane. The pixel count results indicate that the plant exhibits an anticipatory behavior in that it starts to close its leaves prior to the light-to-dark transition. Furthermore, the optical flow results indicate that each compound leaf show different behavior depending on the whereabouts in the circadian rhythm cycle. This suggests that a complex regulating structure lies behind the plant’s response to different illumination regimes.