SignAI: A Visuo-Linguistic Framework for ASL

Inspiration

I was inspired to build SignAI to address communication barriers for the deaf community, creating a tool that interprets American Sign Language in real-time using AI to foster inclusivity and accessibility.

What I Learned

I gained deep insights into computer vision and deep learning, mastering tools like TensorFlow and OpenCV. I also learned how to optimize models for real-time performance and tackle challenges in sign language interpretation.

How I Built It

SignAI was built using Python, TensorFlow, and OpenCV, with SSD MobileNet V2 for efficient hand gesture detection. I trained the model on ASL datasets, improving accuracy through iterative prototyping and testing.

Challenges I Faced

A major challenge was handling low-light conditions and complex backgrounds, which caused false positives in detection. I also worked to improve the model’s speed, achieving up to 38 fps, while maintaining high accuracy.

Built With

• opencv
• python
• ssdmobilenev2
• tensorflow