mkblbj/lpr

LPR

lpr（license plate recognition）

构建及安装

1. 源码准备

• 包括deps，include，models，src文件夹

2. 准备环境

• 安装opencv4.0及以上， freetype库
• 安装cmake3.0以上版本，支持c++11的c++编译器，如gcc-6.3

3. 编译安装

mkdir build
cd build
cmake ..
make install

使用及样例

1.使用MTCNN检测

• 代码样例

void test_mtcnn_plate(){
    pr::fix_mtcnn_detector("../../models/float", pr::mtcnn_float_detector);
    pr::PlateDetector detector = pr::IPlateDetector::create_plate_detector(pr::mtcnn_float_detector);

    pr::fix_lpr_recognizer("../../models/float", pr::float_lpr_recognizer);
    pr::LPRRecognizer lpr =  pr::float_lpr_recognizer.create_recognizer();
    Mat img = imread("../../image/plate.png");

    ncnn::Mat sample = ncnn::Mat::from_pixels(img.data, ncnn::Mat::PIXEL_BGR, img.cols, img.rows);
    std::vector<pr::PlateInfo> objects;
    detector->plate_detect(sample, objects);
    lpr->decode_plate_infos(objects);

    for (auto pi : objects)
    {
        cout << "plate_no: " << pi.plate_color << pi.plate_no << " box:" << pi.bbox.xmin << ","
        << pi.bbox.ymin << "," << pi.bbox.xmax << "," << pi.bbox.ymax << "," << pi.bbox.score << endl;
    }
}

2.使用LFFD检测

• 代码样例

void test_lffd_plate()
{
    pr::fix_lffd_detector("../../models/float", pr::lffd_float_detector);
    pr::PlateDetector detector = pr::IPlateDetector::create_plate_detector(pr::lffd_float_detector);

    pr::fix_lpr_recognizer("../../models/float", pr::float_lpr_recognizer);
    pr::LPRRecognizer lpr =  pr::float_lpr_recognizer.create_recognizer();
    Mat img = imread("../../image/plate.png");

    ncnn::Mat sample = ncnn::Mat::from_pixels(img.data, ncnn::Mat::PIXEL_BGR, img.cols, img.rows);
    std::vector<pr::PlateInfo> objects;
    detector->plate_detect(sample, objects);
    lpr->decode_plate_infos(objects);

    for (auto pi : objects)
    {
        cout << "plate_no: " << pi.plate_color << pi.plate_no << " box:" << pi.bbox.xmin << ","
             << pi.bbox.ymin << "," << pi.bbox.xmax << "," << pi.bbox.ymax << "," << pi.bbox.score << endl;
    }
}

3.使用SSD检测

• 代码样例

void test_ssd_plate()
{
    pr::fix_ssd_detector("../../models/float", pr::ssd_float_detector);
    pr::PlateDetector detector = pr::IPlateDetector::create_plate_detector(pr::ssd_float_detector);

    pr::fix_lpr_recognizer("../../models/float", pr::float_lpr_recognizer);
    pr::LPRRecognizer lpr =  pr::float_lpr_recognizer.create_recognizer();
    Mat img = imread("../../image/manys.jpeg");

    ncnn::Mat sample = ncnn::Mat::from_pixels(img.data, ncnn::Mat::PIXEL_BGR, img.cols, img.rows);
    std::vector<pr::PlateInfo> objects;
    detector->plate_detect(sample, objects);
    lpr->decode_plate_infos(objects);

    for (auto pi : objects)
    {
        cout << "plate_no: " << pi.plate_color << pi.plate_no << " box:" << pi.bbox.xmin << ","
             << pi.bbox.ymin << "," << pi.bbox.xmax << "," << pi.bbox.ymax << "," << pi.bbox.score << endl;
    }
}

4.使用量化模型

• 代码样例

void test_quantize_mtcnn_plate(){
    pr::fix_mtcnn_detector("../../models/quantize", pr::mtcnn_int8_detector);
    pr::PlateDetector detector = pr::IPlateDetector::create_plate_detector(pr::mtcnn_int8_detector);

    pr::fix_lpr_recognizer("../../models/quantize", pr::int8_lpr_recognizer);
    pr::LPRRecognizer lpr =  pr::int8_lpr_recognizer.create_recognizer();
    Mat img = imread("../../image/plate.png");

    ncnn::Mat sample = ncnn::Mat::from_pixels(img.data, ncnn::Mat::PIXEL_BGR, img.cols, img.rows);
    std::vector<pr::PlateInfo> objects;
    detector->plate_detect(sample, objects);
    lpr->decode_plate_infos(objects);

    for (auto pi : objects)
    {
        cout << "plate_no: " << pi.plate_color << pi.plate_no << " box:" << pi.bbox.xmin << ","
             << pi.bbox.ymin << "," << pi.bbox.xmax << "," << pi.bbox.ymax << "," << pi.bbox.score << endl;
    }
}

LPR （License Plate Recognition）

LPR is a quasi-commercial grade license plate recognition library for mobile, with NCNN as the inference backend, using DNN as the core of the algorithm, supporting a variety of license plate detection algorithms, supporting license plate recognition and license plate color recognition.

Features

• Ultra-lightweight, the core library relies only on NCNN and supports model quantization
• Multi-detection, support SSD, MTCNN, LFFD and other target detection algorithms
• High accuracy, LFFD target detection reaches 98.9 in CCPD detection AP, license plate recognition reaches 99.95%, comprehensive recognition rate over 99%
• Easy to use, only need 10 lines of code to complete license plate recognition
• Easy to expand, can quickly expand all kinds of detection algorithms

Algorithms Flow

Build and Installation

1. Source code preparation

• incloud deps，include，models，src Folders

2. Environment Preparation

• opencv4.0 and above, freetype
• cmake3.0 and above, c++ compiler that supports c++11, such as gcc-6.3

3. Compilation and installation

mkdir build
cd build
cmake ..
make install

Usage and samples

1.Detection with MTCNN

• Sample Code

void test_mtcnn_plate(){
    pr::fix_mtcnn_detector("../../models/float", pr::mtcnn_float_detector);
    pr::PlateDetector detector = pr::IPlateDetector::create_plate_detector(pr::mtcnn_float_detector);

    pr::fix_lpr_recognizer("../../models/float", pr::float_lpr_recognizer);
    pr::LPRRecognizer lpr =  pr::float_lpr_recognizer.create_recognizer();
    Mat img = imread("../../image/plate.png");

    ncnn::Mat sample = ncnn::Mat::from_pixels(img.data, ncnn::Mat::PIXEL_BGR, img.cols, img.rows);
    std::vector<pr::PlateInfo> objects;
    detector->plate_detect(sample, objects);
    lpr->decode_plate_infos(objects);

    for (auto pi : objects)
    {
        cout << "plate_no: " << pi.plate_color << pi.plate_no << " box:" << pi.bbox.xmin << ","
        << pi.bbox.ymin << "," << pi.bbox.xmax << "," << pi.bbox.ymax << "," << pi.bbox.score << endl;
    }
}

• Result Example:

2.Detection with MTCNN

• Sample Code

void test_lffd_plate()
{
    pr::fix_lffd_detector("../../models/float", pr::lffd_float_detector);
    pr::PlateDetector detector = pr::IPlateDetector::create_plate_detector(pr::lffd_float_detector);

    pr::fix_lpr_recognizer("../../models/float", pr::float_lpr_recognizer);
    pr::LPRRecognizer lpr =  pr::float_lpr_recognizer.create_recognizer();
    Mat img = imread("../../image/plate.png");

    ncnn::Mat sample = ncnn::Mat::from_pixels(img.data, ncnn::Mat::PIXEL_BGR, img.cols, img.rows);
    std::vector<pr::PlateInfo> objects;
    detector->plate_detect(sample, objects);
    lpr->decode_plate_infos(objects);

    for (auto pi : objects)
    {
        cout << "plate_no: " << pi.plate_color << pi.plate_no << " box:" << pi.bbox.xmin << ","
             << pi.bbox.ymin << "," << pi.bbox.xmax << "," << pi.bbox.ymax << "," << pi.bbox.score << endl;
    }
}

• Results Example:

3.Detection with SSD

• Sample Code

void test_ssd_plate()
{
    pr::fix_ssd_detector("../../models/float", pr::ssd_float_detector);
    pr::PlateDetector detector = pr::IPlateDetector::create_plate_detector(pr::ssd_float_detector);

    pr::fix_lpr_recognizer("../../models/float", pr::float_lpr_recognizer);
    pr::LPRRecognizer lpr =  pr::float_lpr_recognizer.create_recognizer();
    Mat img = imread("../../image/manys.jpeg");

    ncnn::Mat sample = ncnn::Mat::from_pixels(img.data, ncnn::Mat::PIXEL_BGR, img.cols, img.rows);
    std::vector<pr::PlateInfo> objects;
    detector->plate_detect(sample, objects);
    lpr->decode_plate_infos(objects);

    for (auto pi : objects)
    {
        cout << "plate_no: " << pi.plate_color << pi.plate_no << " box:" << pi.bbox.xmin << ","
             << pi.bbox.ymin << "," << pi.bbox.xmax << "," << pi.bbox.ymax << "," << pi.bbox.score << endl;
    }
}

• Results Example:

4.Detection with quantitative models

• Sample Code

void test_quantize_mtcnn_plate(){
    pr::fix_mtcnn_detector("../../models/quantize", pr::mtcnn_int8_detector);
    pr::PlateDetector detector = pr::IPlateDetector::create_plate_detector(pr::mtcnn_int8_detector);

    pr::fix_lpr_recognizer("../../models/quantize", pr::int8_lpr_recognizer);
    pr::LPRRecognizer lpr =  pr::int8_lpr_recognizer.create_recognizer();
    Mat img = imread("../../image/plate.png");

    ncnn::Mat sample = ncnn::Mat::from_pixels(img.data, ncnn::Mat::PIXEL_BGR, img.cols, img.rows);
    std::vector<pr::PlateInfo> objects;
    detector->plate_detect(sample, objects);
    lpr->decode_plate_infos(objects);

    for (auto pi : objects)
    {
        cout << "plate_no: " << pi.plate_color << pi.plate_no << " box:" << pi.bbox.xmin << ","
             << pi.bbox.ymin << "," << pi.bbox.xmax << "," << pi.bbox.ymax << "," << pi.bbox.score << endl;
    }
}

• Results Example:

Follow-Up works

• Add better algorithm support
• Optimize the model, support more license plate types, currently support common license plate recognition
• Optimize the model, higher accuracy
• Performance evaluation

Reference

1. light-LPR
2. NCNN
3. LFFD
4. CCPD Chinese license plate dataset, reaching 2 million samples
5. HyperLPR