Handcheck Detection
High-volume automated detection of checkboxes, ticks, and crosses in forms, replacing manual review of handcheck fields at scale.
The Business Problem
Forms contain checkboxes, ticks, and crosses that need detection and classification at scale. Generic object detectors were either too slow for high volume or inaccurate on small, similar-looking marks.
Manual review of handcheck fields did not scale, and downstream logic depended on reliable detection of marked vs unmarked fields.
The Technical Solution
I built a custom object detection model with MobileNetV3 to identify handcheck marks: checkboxes, ticks, and crosses. The lightweight architecture is tuned for fast inference at high document volume.
The model generalizes across diverse form layouts with labeled training data spanning multiple form types.
The Scalability Factor
Deployed on AWS with Docker as a containerized service in the form processing pipeline. Low latency per document supports high-volume batch processing without bottlenecks.
Business Impact
Handcheck marks detected at 89% accuracy at scale; integrated into the form processing pipeline.
High-volume processing supported with low latency per document.