Introduction

Evo is core engine library to detect and recognize vehicle's license plates. This engine uses Deep Learning technology and parellel processing for better accuracy and performance. For Deep Learning, several inherent Deep Neural Network(DNN for brevity) models were developed and trained continuously. As Deep Learning's sevaral advantages, training the sample images suffices to recognize new or abnormal license plates rather than revising engine's source code. In addition, upgrading the DNN models and retraining them are enough to increase engine's accuracy and performance without modification of code logic.

It takes longer execution time for DNN models to be run than conventional image processing. To overcome this problem, Evo engine uses parallel processing to reduce latency and supports several kinds of devices to increase throughput on which the DNN models can be run. In case of parallel processing, the engine instance processes each input image using parallel mechanism inherent to the device which is assigned to it. For the second case, there are several kinds of devices such as CPU, GPU and so forth on which the DNN models can be run, which can be assigned to different engine instances.
For detailed description, refer to DNN Details.

Workflow Overview

The following block diagram shows Evo engine's overall processing workflow.