Expand-A-Conda

• 

  Image Segmentation using NVIDIA MIT B-5 model

Inspiration

• Moved by firsthand accounts in post-earthquake Turkey, inspired our hackathon project. In response to stories of friends lost and agonizing waits for rescue, we created the Expand-a-Conda, an innovation born from real-life experiences, designed to swiftly detect and locate individuals buried in debris. This initiative combines technology with empathy, offering a lifeline in the aftermath of disasters.

What it does

• Expand-a-conda is designed to detect people buried in debris using image segmentation and AI-powered path finding while also incorporating soft robotics by building a cheap and indestructible vine robot.

How we built it

• We used a finetuned version of the NVIDIA MIT-B5 model to detect humans trapped in rubble. We also used 3D printing technologies to mount a camera on our indestructible vine robot and designed it to navigate through inaccessible terrains. The video feed from the camera is reverse-proxied through Cloudflare and accessed by Google Cloud for processing.

Challenges we ran into

• The only challenge we faced and hope to improve in a later version of this project is the computing power. We have implemented our project with limited resources, while also maintaining the accuracy. Right now, we are getting a low frame rate which can be solved by using local GPUs and higher computing power.

Accomplishments that we're proud of

• We designed a solid proof of concept and developed a fully-functional prototype to demonstrate its feasibility.

What we learned

• Dealing with limited computing power challenged us to find innovative solutions. We had to optimize our algorithms and processes to achieve the desired functionality without compromising on accuracy. This helped us to learn how to hone our problem-solving skills under real-world constraints.

What's next for Expand-A-Conda

• We plan to achieve higher computing power to get a better frame rate, resulting in faster and accurate detection of people trapped in such situations. We also plan to develop remote control for the robot and deploy them in swarms to cover larger search areas. Apart from this, we also hope to incorporate a user identification system using opencv library which can also help is easy identification of the victims.

Built With

• google-cloud
• python
• pytorch
• vision-api