Neural network based auto tag identification system

The growth in automobile usage for transportation requires the application of technology to enhance the traffic flow as well as safety. Some of the needed improvements are in the areas of automated toll collection at toll booths, monitoring of traffic flow and tracking of stolen vehicles. We intend to accomplish them with the identification of auto tags at toll collection sites and interfacing the tag information to a database through a computer network. The tag information of an automobile is identified in three stages. In the first stage, the image of the auto tag is captured by an electronic eye under a well-lit condition at the entry point of the toll collection. Next, the rectangular tag area is extracted from the color image with image filters. Subsequently, the bounding rectangle of each character of the tag image is partitioned. Then each segment is normalized to the standard orientation and size. Also, the color of the character segment is transformed into a black and white (B&W) image. A feedforward neural net has been trained to distinguish a fixed set of B&W image characters (alphabets and numerals). In the second stage, each character segment image of the auto tag is fed to the neural net and the classification results are grouped together. This identified tag value is stored in a database with the corresponding time-stamp and this information is shared through a computer network. The simplicity of the approach relies on the standardization of the image and converting each character segment image into 1-bit B&W pixels. This facilitates in the reduction of the input vector size for the neural net