EdgeSight

• 

  cover.png EdgeSight: computer vision that runs entirely in your browser, with no servers and no uploads.

• 

  why.png Why EdgeSight matters: it keeps photos private, needs no servers, works offline and makes image processing easy to learn.

• 

  -architecture.png A fully serverless setup. The static site is stored in Amazon S3 and delivered through CloudFront, and all processing .

• 

  filters.png Core vision filters running live on the user’s device: grayscale, blur and Sobel edge detection, all in real time.

Inspiration

Most computer vision tools need a GPU server, a paid API or fast internet. Students, schools, small clinics and farms in low-connectivity areas often have none of these. Sending photos to a remote server also raises privacy concerns.

What it does

EdgeSight is a static web app that lets anyone run core computer vision operations on their own photos. These are grayscale, blur, Sobel edge detection and thresholding. Everything is processed on the user’s device, so nothing is uploaded. Once the page has loaded, it works offline.

How we built it

HTML, JavaScript, Canvas API, Amazon S3, Amazon CloudFront HTML, CSS and JavaScript, with image processing on the canvas using convolution kernels (the same maths OpenCV uses for Sobel and blur). Hosted on AWS as static files: Amazon S3 for storage and Amazon CloudFront for fast global delivery. No backend, database or physical server to manage.

Challenges: Keeping the processing fast enough on phones and low-end laptops, and keeping the app small enough to load quickly on slow connections.

Challenges we ran into

: Keeping the processing fast enough on phones and low-end laptops, and keeping the app small enough to load quickly on slow connections.

Accomplishments that we're proud of

EdgeSight is a static web app that lets anyone run core computer vision operations on their own photos. These are grayscale, blur, Sobel edge detection and thresholding. Everything is processed on the user’s device, so nothing is uploaded. Once the page has loaded, it works offline.

What we learned

Most computer vision tools need a GPU server, a paid API or fast internet. Students, schools, small clinics and farms in low-connectivity areas often have none of these. Sending photos to a remote server also raises privacy concerns.

What's next for EdgeSight

Swap in OpenCV.js for more advanced features such as contour and object detection. Plant leaf checks for farmers. Document edge detection for phone scanning. A classroom mode for teaching image processing.

Built With

• 23.
• accessibility
• amazon-cloudfront
• amazon-web-services
• canvas-api
• client-side
• convolution
• css3
• edge-computing
• edge-detection
• education
• html5
• image-processing
• javascript
• low-bandwidth
• privacy
• responsive-design
• serverless
• sobel-operator
• static-website
• web-app