The project aims at developing a detector running on Raspberry pi as a low cost solution for monitoring endangered species such as tigers. The model will be a variant of the squeeze net which takes 4.8 MB hardware space. Squeezenet is a derivative of ALexnet and is used for classification. I am trying to implement image detection using the classification head of the squeezenet and adding a bounding box regressor that regresses over the grid patches in which the image would be divided to find the animal and locate it. The purpose of detection is to detect each instance of the tiger and to differentiate them. One of the biggest problems in tiger population counting is to differentiate between the individual animals. This is currently done on the basis of tigers paw marks and other differentiable properties. However with the use of computer vision and deep learning I feel we can make much better predictions without too much human intervention which is necessary to let these animals reside peacefully in the small space that they have in the woods. The problem at hand does not require a lot of processing power and considering the size and time efficiency of models like squeezenet I am trying to figure out how to bring the effectiveness and accuracy of neural networks to low cost solution such as raspberry pi. Monitoring the endangered species is one cause that I am particularly interested in as these animals are a vert important part of the foodchain.

I would be using tensorflow for the implementation. Some of the links that I am finding useful are:

https://github.com/samjabrahams/tensorflow-on-raspberry-pi

https://svds.com/tensorflow-image-recognition-raspberry-pi/