Wyniki 1-5 spośród 5 dla zapytania: authorDesc:"Grzegorz Szwoch"

Performance evaluation of parallel background subtraction on GPU platforms DOI:10.15199/13.2015.4.4


  Automated, on-line video content analysis is the current trend in surveillance systems. Such solutions evolved from trivial motion detector algorithms into sophisticated methods, such as multiple object tracking, unattended object detection, crowd monitoring, etc [1]. Algorithms performing video content analysis in these systems usually need high processing power. Moreover, modern video surveillance systems utilize high resolution cameras which puts even more load on the processing stations. An example of such an algorithm is background subtraction (BS), an underlying part of object detection and tracking methods. The aim of this algorithm is to divide video image pixels into classes: foreground (moving objects), background (static scene) and optionally also shadows. Proper realization of this goal usually requires employing a statistical model. A method most often used for this task is the Gaussian Mixture Models (GMM) approach which was proposed for BS in video by Stauffer and Grimson [2] and latter extended by Zivkovic [3] and others. The BS procedure operates on the pixel level and performs a number of computationally intensive operations on each image pixel and its background model. For high resolution cameras, the number of pixels to process is so large that it is not possible to perform on-line (live) BS using GMM on the CPU. At the same time, the GMM algorithm is parallel in nature (even so-called “embarrassingly parallel") because each pixel is processed independently from the others. Therefore, more than one pixel may be processed at the same time using a parallel computing device [4]. With modern CPUs, computations may be performed in a parallel manner using systems such as OpenMP [5], but the number of processing cores is very limited, so the processing power is still not enough for on-line video analysis. In the recent years, parallel computing on Graphics Processing Units (GPUs) has become a widely used [...]

Estimation of object size in the calibrated camera image

Czytaj za darmo! »

Video surveillance systems are becoming a common tool for improving the public security. However, the efficiency of these systems is limited by a fact that a human operator has to observe constantly a large number of monitors for detection of events occurring in the images captured by cameras. One method of improving the efficiency of the discussed systems is introduction of the automatic tool that assists the human operator by automatically analysing the camera images and informing the operator about important events that were detected [1]. However, accurate detection of events in the camera images requires that the detected moving objects are assigned to proper classes and this classification requires that the physical size of the tracked object (not the size in image pixels[...]

Resolving conflicts in object tracking for automatic detection of events in video


  In modern times, video monitoring systems become increasingly complex, consisting of numerous cameras and monitors. It is not possible that the system operator notices every important event that occurs in the monitored areas. Therefore, an automatic tool that analyses images from all cameras and detects important security threats, would be able to assist the operator, noticing him on the detected events and thus improving the level of public security. Such a system is currently under development by the authors of this paper [1, 5]. Performance of automatic event detection in video from monitoring cameras depends on accuracy of the low-level image analysis algorithms, such as detection and tracking of moving objects. The first low-level procedure is usually related to separation of moving object from the static background in each image frame from the camera. The authors use a method based on background modeling as proposed by Stauffer and Grimson [4], which models statistical properties of the background with a mixture of Gaussian models (GMM). The results of background modeling are processed in order to remove shadow pixels (using color and brightness distortion measures) and finally they are cleaned using morphological operations (removing noise, closing holes, etc.). The next processing stage is related to tracking movement of detected objects in successive camera images. For this task we use an effective approach based on Kalman filters [6], although various alternative methods may be found in the literature. However, none of the tracking algorithms define a method of assigning trackers to objects detected in the frame and resolving ambiguous relations between them. Therefore, a procedure for resolving such tracking conflicts was developed, the algorithm is described in the next sections of the paper. The results of object detection and tracking may be used for automatic event detection. The event detector uses a set of d[...]

ZASTOSOWANIA DRONÓW I SENSORÓW WIZYJNYCH I AKUSTYCZNYCH DO ZDALNEJ DETEKCJI I LOKALIZACJI OBIEKTÓW I ZDARZEŃ DOI:10.15199/59.2016.8-9.75


  W referacie przedstawiono wybrane sensory akustyczne i wizyjne i propozycje ich zastosowania do wykrywania i lokalizacji obiektów i zdarzeń z pokładu drona. Opisano pokrótce zastosowane algorytmy analizy strumieni, przedstawiono wyniki badań stworzonych prototypów i metod, zaimplementowanych na wydajnych układach GPU Abstract: The paper presents acoustic and visual sensors and their application to detection and localization of objects and events on board of unmanned aerial vehicles. Developed algorithms and methods are described and evaluated, and power consumption and performance are reported. Several scenarios are proposed. Słowa kluczowe: Analiza dźwięku, analiza obrazu, sensory Keywords: Image analysis, sensors, sound analysis 1. WSTĘP Bezzałogowe statki powietrzne (ang. unmanned aerial vehicle - UAV), potocznie nazywane dronami, zyskują w ostatnim czasie dużą popularność. Wiele z nich jest wyposażonych w kamery wideo i umożliwią strumieniowanie obrazu do aplikacji klienta. Uzupełnienie w inne typy sensorów pozwoli na automatyczne wykrywanie i śledzenie obiektów i zdarzeń związanych z zagrożeniem. Na pokładzie drona zamontowane mogą być kamery, czujniki podczerwieni, skanery laserowe, odległościomierze, sensory akustyczne i inne. Z ich pomocą prowadzone może być lokalizowanie i monitorowanie zjawisk i procesów, nawigowanie i autonomiczne poruszanie się w środowisku. Zastosowania obejmują ochronę granic, zarządzanie kryzysowe, monitorowanie infrastruktury krytycznej, wykrywanie i zwalczanie skutków pożarów, klęsk żywiołowych i katastrof ekologicznych, potencjalnie wspomagając człowieka w niebezpiecznych lub czasochłonnych zadaniach. W prezentowanych badaniach sprawdzono wydajność obliczeniową i skuteczność działania wybranych algorytmów analizy strumieni wizyjnych i akustycznych. Zakłada się, że dane przekazywane są z czujników siecią bezprzewodową do naziemnej jednostki obliczeniowej. Zaproponowano i przetestowan[...]

 Strona 1