This technology is a detection method and device that identifies and merges individual character regions in road driving footage through edge enhancement, noise reduction, and connected component analysis to select license plate candidates, then determines the final license plate area using a Support Vector Machine (SVM) based on edge density and color feature vectors for each sub-region.
As the sharing of street view and dashcam footage has increased, privacy infringement due to exposed license plates has become a concern, yet manually blurring the massive volume of daily footage is practically impossible. Existing detection methods have limitations, as they are vulnerable to variations in license plate position, size, tilt, and lighting conditions, as well as confusing elements like road signs and advertisements.
This technology significantly reduces the number of candidate regions by applying conditions for character height, center lines, and spacing, as well as license plate aspect ratios, before using a trained classification model for identification, thereby improving both detection rates and processing speed. It can be applied to automatic de-identification for dashcam sharing platforms, parking management, and toll systems, allowing for rapid license plate detection even under diverse shooting conditions.
N/A
