Towards multiple license plate localization in Indian conditions: An edge density based approach

Vehicles are generally identified by their license plates. Locating the exact area containing the license plate is an important step in any license plate recognition system. This work describes an Edge Density Based approach for locating the single/multiple license plates present in the captured frame. The proposed approach is divided into two stages: Pre-processing and Exact detection. Pre-processing involves RGB to Grayscale conversion and Contrast Enhancement. Contrast Enhancement is employed to enable the algorithm to detect license plates at night and in low light conditions. In the second stage after edge detection, Edge Density Based Region Growing is applied to highlight the regions with dense vertical edges that may contain the license plate. The highlighted regions are extracted one-by-one through connected component analysis. These are then subjected to six-different filtering tests to check whether they are license plates or not. The candidates which qualify all the tests are returned as Exact Detected License Plates. To show the effectiveness of the proposed approach a detailed evaluation is done on a database comprising of 810 images captured from India and abroad. This database contains vehicles having license plates of variable size, shape and color captured in different environmental conditions. The proposed approach detects license plates with 91% of accuracy.