Real time recording and monitoring of human activities and animal movements in protected areas.
- 0 Collaborators
A cost-effective and intelligent end to end solution with Android and web services to stop poaching. Deep Learning is the core of the solution and whatever done on the project is state of the art. The project won an award in Smart India Hackathon 2018 but real life implementation yet to be done. ...learn more
Project status: Under Development
Mobile, Artificial Intelligence
Overview / Usage
Deep Learning based solution to tackle poaching. Several features like - object detection, Image classification, audio classification etc are being heavily used to track activities in a wildlife sanctuary. The solution is state of the art, cost-effective and is tested and ready. This will help stop poaching and will help in tracking animals in wildlife.
Methodology / Approach
An affordable, scalable and state of the art, end to end solution for tackling poaching and tracking animals/humans in protected areas.The whole idea comprises of three parts:
- Android
- Web
- Intelligence (Deep Learning)
The idea is that the Rangers will have a lightweight android app with them and they will be able to monitor/report activities in wildlife. The android app ui has a map, showing real-time location of all the rangers in black and displays some other features like temperature, humidity etc. The UI also has a red button, so that the rangers can press at any time in case they detect/suspect any poaching activity. As soon as they press the red button, the longitude and latitude of the ranger pressing the button will be sent to the web server and the server will update each and every rangers app showing the shortest path from each and every rangers location to the location in DANGER(in RED Color). The shortest path will be found just like that of UBER. More information on finding the shortest path is noted down in the GitHub link. The whole Idea is large and involved many parts, including Raspberry Pi, Camera configuration etc to meet the ends. To detect and track animals YOLOv2 trained on COCO is used for object detection, ResNet50 trained on Imagenet for Image Classification, Audio in frequency image forms to track animals via sensors was trained using a Convolutional Neural network and classified into different categories of animals. Image Processing and Predictive analysis based solution to predict and find shortest path was established. Detailed description in the GitHub link.
Technologies Used
Darkflow-Darknet
ResNet50-TensorFlow
CNN
Python
OpenCV
Android
PHP, NodeJS for Web Services