Aleksander Lamża

Depth Estimation in Image Sequences in Single-Camera Video Surveillance Systems

Depth estimation plays a key role in numerous applications, including video surveillance, target tracking, robotics or medicine. The standard method of obtaining depth information is to use stereovision systems, which require at least two cameras. In some applications this is a big hindrance because of dimensions, costs or power consumption. Therefore, there is a need to develop an efficient method of depth estimation that can be applied in single-camera vision systems. In the literature several techniques to accomplish this task can be found. However, they require either modification of the camera (in the mirror-based methods) or changing the parameters of the lens (in the focus-based methods). In this paper a new method based on image sequences from cameras with standard fixed-focal-length and fixed-focus lens is presented.

New Efficient Method of Digital Video Stabilization for In-Car Camera

The problem of image stabilization in video sequences is important in many applications, including cameras mounted on vehicles. This type of vision systems can be used in vehicles that are the equipment of police and other road services as well as in driving assist systems. The stability of video aquired while driving directly affect the quality resulting in performance of further processing. In this paper, the new approach to digital video stabilization was presented. The proposed method is based on block-matching algorithm, namely Gray-Coded Bit Matching Plain. Modification of the algorithm improves performance by reducing the size of the analyzed block. This was achieved by the introduction of additional information of acceleration in the vertical axis aquired from accelerometer coupled with camera.

Methods for Face Localization in Static Colour Images with an Unknown Background

This paper analyses the practical application of the methods for face and head localization in colour images with varying background. The Haar Cascade Classifier and Local Binary Pattern were selected as basic methods because of their high efficiency in this type of applications. The results obtained in the test set of images prompted the authors to choose the Haar Cascade Classifier. This method was then implemented in the face detection module, which is part of a comprehensive system for determining the forbidden regions.

A metropolitan assistive system for disabled and the elderly

In the article the general concept of system supporting patients, disabled and the elderly was presented. The assumptions of the system is to improve the quality of life and comfort those who are not able to function without assistance. The system consists of network infrastructure and mobile devices used by beneficiaries of support. Developing a detailed specification requires the cooperation of specialists in many fields, especially doctors, therapists, senior caregivers and disabled people themselves, because it is of critical importance to design the system according to the needs of the potential system users. Despite all the problems, it seems that the development of the abovedescribed systems is a job worth doing.

Dynamics of the Clusterization Process in an Adaptative Method of Image Segmentation

In this paper a new method of automatic image segmentation is presented. In proposed approach a region merging method is applied, where pixels or pixel groups are merged to bigger areas. In the proposed method of the region merging seed settings, segmentation parameter definition, and number of outcoming areas determining can be ommited. Homogeneous clusters (consisting of continuous groups of pixels) are the part of the growth process. The clusters are merged when the homogeneousness condition is fulfilled. The threshold value changes during segmentation process, appropriate to the changeable conditions.

The Clusterization Process in an Adaptative Method of Image Segmentation

In the article a new method of automatic image segmentation is presented. The aim was to eliminate the necessity of defining the number of outcome areas. Homogeneous areas take part in the growth process. The areas merge when the homogeneousness condition is fulfilled. The threshold value changes during the segmentation process, fitting the changeable conditions.