|
Project properties |
|
| Title | Automatic Wild Fauna Video-Monitoring in Outdoor Facility of Poultry Flock |
| Group | Information Technology Group |
| Project type | thesis |
| Credits | 24-39 |
| Supervisor(s) | Cagatay Catal |
| Examiner(s) | Cagatay Catal |
| Contact info | bedir.tekinerdogan@wur.nl
cagatay.catal@wur.nl |
| Begin date | 2020/05/01 |
| End date | 2024/08/31 |
| Description | Wild birds are a thread for free-range poultry flocks, because they can have direct or indirect contact with chickens in the outdoor facility, enabling transmission of contagious animal diseases like bird flu causing severe outbreaks leading to complete depopulation of the farm. To monitor which, where, when and how often wild fauna visit the outdoor facility, a total of eight 1.3 Mpx TruVision IP 1/3" CMOS video-cameras were installed by researchers of Wageningen Bioveterinary Research. Cameras were connected to a TruVision NVR10 network video recorder with HDMI/VGA video-output and a 4TB hard disk for storage. The cameras were equipped with IR LEDs enabling night-recording. Recording was done at a speed of 2 frames/sec, 24 hr/day, 7 days/week, enabling recording of about 30 observation days per recorder before a recorder had to be refreshed. From these video recordings it could be concluded which type of wild species are visiting when, for how long, and with what numbers the outdoor facility during the year. Because these videos have to be inspected by the human eye being very labor intensive, this hinders useful applications of video cameras for animal monitoring. However, smart cameras are used more and more for automatic object detection. Therefore, it should be feasible to use these WBVR video-data to build an automatic monitoring system with classification and localization of the relevant types of wild animals.
To provide an working and applicable automatic wild animal video-monitoring, an artificial neural network should be trained to classify and localize the relevant types wild animals and distinguish them from the chicken flock. For this analysis, the video-recordings of WBVR should be used in combination with general available images of the relevant wild types. To enable this, the video recordings of WBVR should be downloaded from the data-recorders and put on easy accessible disks space. The trained object detection network should be importable in the software packages R and Mathematica to facilitate the detection of visiting moments and number of visits for each type of wild animal. |
| Used skills | |
| Requirements | Software Engineering, Programming in Python, Machine Learning |