Smart surveillance system using edge-devices for wildlife preservation in animal sanctuaries
||Smart surveillance system using edge-devices for wildlife preservation in animal sanctuaries|
||Magnus Malmström <email@example.com>|
||2023-02-01 – 2023-08-01|
Today, camera surveillance is widely used and thanks to advanced algorithms and machine learning for object-detection, modern cameras can achieve a higher understanding of what is seen and communicate this to a user. This leads to there being now smart and sophisticated surveillance systems all over the world with different areas of use.
These systems can often be found in urban environments and areas with a lot of people passing through. There are, however, sparsely populated and remote areas that could benefit from the use of information provided by smart surveillance systems. One example is animal sanctuaries that provide shelter for exposed and endangered species. These areas are often patrolled by park rangers to ensure the safety of the animals and protect the area from unauthorized people. The ranger's work could become more effective by the use of smart surveillance systems that sends information to the rangers in real-time about activities in the sanctuary. This project will focus on both animals in the African savanna as well as in the Swedish forest.
Edge-devices are microcomputers that are generally used to collect, process, and forward information. These devices are being used to a greater extent in time with smart units like smart cars, speakers, homes, and cameras that are becoming more popular. Edge-devices with cameras as sensors can be used to create a smart surveillance system in an environment where a traditional solution is not available. The edge-devices can be independent when it comes to power supply and connection, making it possible to place them in an animal sanctuary for surveillance in the area.
To make the use of classification algorithms on edge-devices possible, it is required for them to be pretrained on a more powerful computer. The classifications algorithms to be used in this project are neural networks of moderate sizes, which should be able to run on edge-devices. Hence, to train the neural networks that should be used to distinguish rhinos from poachers it would be beneficial to use the GPU resources, such as those found at the cluster Berzelius. It would save time such as it would be possible to try out an increasing number of settings of hyperparameters for the networks.
The project is a master thesis project which is part of the Ngulia project. The Ngulia project has been running at Linköping University since 2014, while the master thesis project has been running since 2019, where this is the fourth iteration of the project. The previous iterations can be found at https://t.co/89YS4ZV3i9 and https://t.co/neXmMY9jYm