A Hierarchical Learning Approach for Finding Multiple Vehicle Number Plate Under Cluttered Background

Abstract

Traffic control is one of the biggest problems faced by the surveillance department in every country. Manual surveillance supported by well-defined rules and effective equipment is a common solution to this problem. But, often traffic personnel fail to isolate and recognize the number plate and hence to penalize the owner of the vehicle who violated the traffic rules due to the speeding vehicle and mounting of number plates at arbitrary parts of the vehicle. Hence, his inability renders the traffic surveillance a challenging research area. The automatic vehicle number plate recognition system provides one of the solutions to this problem but constrained by various limitations. The most challenging aspect is to detect the number plates itself present in an image. The presence of multiple vehicle number plates and cluttered background makes our work different from earlier approaches. A single step process may not be able to detect all the number plates in an image, hence we propose Hierarchical Filtering (HF) approach which employs several transformation functions on the input image. The proposed HF models the characteristics of vehicle number plates and label them by fitting the Logistic Regression model. The proposed method is able to detect all the number plates present in an image but at the expense of some non-number plate regions in the form of a minimum bounding rectangle. The proposed model is tested on a wide range of inputs, and the results are profiled in terms of precision and recall measures.