AnimEye animal detection for road safety

Road safety is generally a great concern for local authorities, therefore they invest a lot in making the roads safer for the driver and pedestrian. However, another concern is preserving the animal live as well, and this is where our product comes in. AnimEye leverages AI technology to solve this problem by detecting animals on the road and sending alerts to local authorities so precautionary actions could be take. We planned our work around 2 iterations, the first one is a simple classifier that classifies whether the input image has animal crossing the road or not, which we have successfully achieved. The second phase would include working on a detector that detects the animals, their location, and counts them then gives an alert with the number of animals on the road and their location. this step is what we hope we will able to achieve in the near future with more time and resources. For the first phase we started with collecting and processing the data from multiple sources: 1- open sources online websites 2- downloading videos from online sources and transforming them into frames so they could be used for model training our train data included 978 images belonging to the 2 classes while our test set included 240 images between the same classes. As for the machine learning / deep learning part we have tried multiple models: 1- building our own model 2- VGG16 pretrained model 3- Resnet50 pretrained model However, the best accuracy we got was from the VGG16 model as follow: train accuracy : 0.91 test accuracy: 0.87 we believe our progress so far is a good proof of concept but understand that there are certainly issues that could be addressed as we work together more knowledge and experience in the future. However, we believe in the greater good that this product could provide for the human life, the economy, and the environment.