A Two-Stage Pipelined Algorithm for Recognition Tasks: Using License Plate Recognition as an Example

Although there has been a lot of research on deep learning, most of them use GPU platform to run deep network models. However, it is less desirable to utilize GPU in real-world scenarios due its relatively high cost and high power consumption. In this paper, we propose a two-stage pipelined algorithm (TSPA) suitable for the FPGA platform to avoid the above-mentioned issues. We also combine OpenCV and GStreamer so that the FPGA platform can achieve real-time performance while maintaining satisfactory accuracy. We choose license plate recognition as an example to demonstrate the feasibility of our proposed approach. We have conducted experiments using the AOLP dataset and the self-collected videos. Our proposed method achieves promising results on these videos.