| Abstract [eng] |
Wild animal monitoring and detection systems are widely used to study animal migration routes and the effects of the environment on them, to protect human-inhabited or dangerous areas from wild animals, and prevent car accidents and other disasters. Monitoring and detection of animals ir also possible with such systems that process images scanned by cameras (subsection 1.2). Systems based on processing of images received by the camera are non-invasive, do not require processing device blocks or stations (subsection 1.4). In this work, algorithms for animal detection will be researched and designed as well as the selected algorithm will be implemented in an embedded system. The system developed by the scientists of Kaunas University of Technology, Faculty of Electricity and Electronics, described in subsection 2.1, was used for the research and development of algorithms for processing images. One of the parts of the system – a thermal imaging camera – was used to capture images, which were processed for further work as described in subsection 2.2. Four algorithms for detecting animal objects in images were analyzed and designed, as depicted in Section 3. According to parameters described in subsection 4.1, different methodologies were investigated and compared; the analysis of techniques and the selection of one methodology for transfer to the embedded system is described in subsection 4.2. The chosen methodology is based on motion capture. The selected methodology was implemented on a microcontroller of the embedded system; some of the steps of the methodology were accelerated by implementing them on the FPGA board. The actualization details of different methodologies, implemented both in microcontrollers and embedded systems, as well as algorithms execution times, were examined and compared in Section 5. A study of the chosen methodology with different photo sets was also carried out. |
