Animal/ Object Identification using deep learning on raspberry pi
- 0 Collaborators
This project aims at bridging the gap between deep learning and IoT. Raspberry Pi comes with very limited computational power and this project was completely based on running a convolutional neural network on raspberry pi by identifying objects real-time from the video feed. ...learn more
Project status: Published/In Market
Overview / Usage
This project solves the problem of using Deep Learning on IoT devices with limited computational capability. Achieving great accuracies for object identification on raspberry pi at a decent processing speed with a little latency.
Methodology / Approach
By using concepts of Transfer Learning, we re-trained the outer layer of inception-v3 model by Google using a custom dataset made using pictures of animals clicked in the town of Ahmedabad. Raspberry Pi supports tensorflow so the whole project was made using native tensorflow scripts. We used Nvidia's 1050Ti GPU to train and the trained model was transferred to Raspberry Pi for testing. Images were provided to R-Pi through external storage and a camera attachment on which we ran our scripts and found the R-Pi to be able to run the CNN scripts efficiently and classify the images at a great speed with a very little latency.
Technologies Used
tensorfloe, keras, NVIDIA GTX 1050Ti, python