An attempt at license plate recognition Using edge

  • Slides: 10
Download presentation
An attempt at license plate recognition Using edge tracing and tangent curves

An attempt at license plate recognition Using edge tracing and tangent curves

Steps in recognition 1. Finding The License Plate 2. Finding And Representing Digits 3.

Steps in recognition 1. Finding The License Plate 2. Finding And Representing Digits 3. Matching

Finding The License Plate • Eliminating all colors on the wrong side of the

Finding The License Plate • Eliminating all colors on the wrong side of the spectrum. Easy done using colormaps to leave only 2 colors white and yellow.

Finding The Digits • Edge Tracing • Color Segmentation How to save the information

Finding The Digits • Edge Tracing • Color Segmentation How to save the information we found? ? ? • Edge chains. • Curve orientation. • Polyline fitting. • Using a method similar to Houghman and Clows, by segmenting the shape into a grid and labeling the type of curve in each segment (concave arc, corner, straight line etc)

Finding the digits • Edge tracing over the filtered plate, going clockwise around the

Finding the digits • Edge tracing over the filtered plate, going clockwise around the pixel edge until back in the starting point. • Averaging the degree over line segments and marking a feature point when a difference in degree between to segments has passed a certain threshold.

Tangent Curve • Marking the curve at each pixel during the trace. • Spline

Tangent Curve • Marking the curve at each pixel during the trace. • Spline interpolation and normalization by arclength • Re-Sampling

Tangent Curve

Tangent Curve

Matching • Give high score for low distance tangent curves. • Needs to be

Matching • Give high score for low distance tangent curves. • Needs to be checked for different orientations. • Longest Common Sub. Sequence – No Help

Results – Needs improvement

Results – Needs improvement

Possible Improvments • Check the tangent curve distance for different orienations. • Better digit-plate

Possible Improvments • Check the tangent curve distance for different orienations. • Better digit-plate segmentation • Applying a grid on the curve features and comparing grids in addition to curve distance to compose a more accurate score