代码拉取完成，页面将自动刷新
本章节在 Lockzhiner Vision Module 上基于 PaddleDet 目标检测类和rcnn文本识别模型，实现了一个车牌识别案例。
车牌识别是一种基于计算机视觉和深度学习的技术，通过图像处理、字符分割和光学字符识别算法，自动提取车辆牌照中的文字与数字信息。该技术可实时识别不同光照、角度和复杂背景下的车牌，广泛应用于智能交通管理、停车场收费系统、电子警察执法等场景。
#include <lockzhiner_vision_module/vision/deep_learning/detection/paddle_det.h>
lockzhiner_vision_module::vision::PaddleDetection();
bool Initialize(const std::string& model_path);
void SetThreshold(float score_threshold = 0.5, float nms_threshold = 0.3);
std::vector<lockzhiner_vision_module::vision::DetectionResult> Predict(const cv::Mat& image);
#include <ncnn/net.h>
ncnn::Net ocr_net;
int load_param(const DataReader& dr);
int load_model(const DataReader& dr);
ncnn::Mat::from_pixels(plate_img.data, ncnn::Mat::PIXEL_BGR, plate_img.cols,
plate_img.rows);
#include <ncnn/net.h>
ncnn::Extractor ex = ocr_net.create_extractor();
lockzhiner_vision_module::vision::PaddleDet detector;
if (!detector.Initialize(argv[1])) {
cerr << "Failed to load detection model: " << argv[1] << endl;
return 1;
}
auto results = detector.Predict(image);
cv::Mat output_image = image.clone();
for (const auto& det : results) {
cv::Rect rect(
static_cast<int>(det.box.x),
static_cast<int>(det.box.y),
static_cast<int>(det.box.width),
static_cast<int>(det.box.height)
);
cv::rectangle(
output_image,
rect,
cv::Scalar(0, 255, 0),
1,
cv::LINE_AA
);
}
cv::imshow("Detection Result", output_image);
ocr_net.load_param(param_path.c_str())
ocr_net.load_model(model_path.c_str())
const float mean[3] = {127.5f, 127.5f, 127.5f};
const float norm[3] = {0.0078125f, 0.0078125f, 0.0078125f};
in.substract_mean_normalize(mean, norm);
string license;
vector<int> preb;
for (int c = 0; c < feat.c; c++) {
const float* data = feat.channel(c);
for (int w = 0; w < feat.w; w++) {
float max_val = -FLT_MAX;
int max_idx = 0;
for (int h = 0; h < feat.h; h++) {
float val = data[w + h * feat.w];
if (val > max_val) {
max_val = val;
max_idx = h;
}
}
preb.push_back(max_idx);
}
}
// 后处理去重
vector<int> valid_labels;
int prev = -1;
for (size_t i = 0; i < preb.size(); ++i) {
if (preb[i] != 67 && preb[i] != prev) {
valid_labels.push_back(preb[i]);
}
prev = preb[i];
}
for (int idx : valid_labels) {
license += plate_chars[idx];
}
自定义函数说明
string RecognizePlate(cv::Mat plate_img);
#include <lockzhiner_vision_module/vision/deep_learning/detection/paddle_det.h>
#include <ncnn/net.h>
#include <lockzhiner_vision_module/edit/edit.h>
#include <chrono>
#include <iostream>
#include <vector>
#include <opencv2/opencv.hpp>
#include "myfontface.h"
using namespace std;
using namespace std::chrono;
// OCR字符集配置
const string plate_chars[68] = {
"京", "沪", "津", "渝", "冀", "晋", "蒙", "辽", "吉", "黑", "苏", "浙",
"皖", "闽", "赣", "鲁", "豫", "鄂", "湘", "粤", "桂", "琼", "川", "贵",
"云", "藏", "陕", "甘", "青", "宁", "新", "0", "1", "2", "3", "4",
"5", "6", "7", "8", "9", "A", "B", "C", "D", "E", "F", "G",
"H", "J", "K", "L", "M", "N", "P", "Q", "R", "S", "T", "U",
"V", "W", "X", "Y", "Z", "I", "O", "-"};
ncnn::Net ocr_net;
bool InitOCRModel(const string& param_path, const string& model_path) {
return ocr_net.load_param(param_path.c_str()) == 0 &&
ocr_net.load_model(model_path.c_str()) == 0;
}
string RecognizePlate(cv::Mat plate_img) {
cv::resize(plate_img, plate_img, cv::Size(94, 24));
ncnn::Mat in = ncnn::Mat::from_pixels(plate_img.data,
ncnn::Mat::PIXEL_BGR,
plate_img.cols,
plate_img.rows);
const float mean[3] = {127.5f, 127.5f, 127.5f};
const float norm[3] = {0.0078125f, 0.0078125f, 0.0078125f};
in.substract_mean_normalize(mean, norm);
ncnn::Extractor ex = ocr_net.create_extractor();
ex.input("input.1", in);
ncnn::Mat feat;
ex.extract("131", feat);
string license;
vector<int> preb;
for (int c = 0; c < feat.c; c++) {
const float* data = feat.channel(c);
for (int w = 0; w < feat.w; w++) {
float max_val = -FLT_MAX;
int max_idx = 0;
for (int h = 0; h < feat.h; h++) {
float val = data[w + h * feat.w];
if (val > max_val) {
max_val = val;
max_idx = h;
}
}
preb.push_back(max_idx);
}
}
// 后处理去重
vector<int> valid_labels;
int prev = -1;
for (size_t i = 0; i < preb.size(); ++i) {
if (preb[i] != 67 && preb[i] != prev) {
valid_labels.push_back(preb[i]);
}
prev = preb[i];
}
for (int idx : valid_labels) {
license += plate_chars[idx];
}
return license.empty() ? "UNKNOWN" : license;
}
int main(int argc, char** argv) {
// 参数验证
if (argc < 4 || argc > 5) {
cerr << "Usage: " << argv[0]
<< " <det_model> <ocr_param> <ocr_model> [image_path]\n"
<< "Example:\n"
<< " Realtime: " << argv[0] << " det_model ocr.param ocr.bin\n"
<< " Image: " << argv[0] << " det_model ocr.param ocr.bin test.jpg\n";
return 1;
}
// 初始化检测模型
lockzhiner_vision_module::vision::PaddleDet detector;
if (!detector.Initialize(argv[1])) {
cerr << "Failed to load detection model: " << argv[1] << endl;
return 1;
}
// 初始化OCR模型
if (!InitOCRModel(argv[2], argv[3])) {
cerr << "Failed to load OCR model: " << argv[2] << " and " << argv[3] << endl;
return 1;
}
MyFontFace myfont;
// 设置文字参数
double font_scale = 0.6;
int thickness = 1;
// 图片处理模式
if (argc == 5) {
cv::Mat image = cv::imread(argv[4]);
if (image.empty()) {
cerr << "Failed to read image: " << argv[4] << endl;
return 1;
}
auto results = detector.Predict(image);
// 可视化并显示结果
cv::Mat output_image = image.clone();
for (const auto& det : results) {
cv::Rect rect(
static_cast<int>(det.box.x),
static_cast<int>(det.box.y),
static_cast<int>(det.box.width),
static_cast<int>(det.box.height)
);
cv::rectangle(
output_image,
rect,
cv::Scalar(0, 255, 0),
1,
cv::LINE_AA
);
}
cout << "\n===== 检测到 " << results.size() << " 个车牌 =====" << endl;
for (size_t i = 0; i < results.size(); ++i) {
const auto& det = results[i];
cv::Rect roi(det.box.x, det.box.y, det.box.width, det.box.height);
roi &= cv::Rect(0, 0, image.cols, image.rows);
if (roi.area() > 0) {
cv::Mat plate_img = image(roi);
cv::imshow("DetectionSeg Result", plate_img);
string plate_num = RecognizePlate(plate_img);
// 左上角偏移
cv::Point text_org(roi.x + 2, roi.y - 2);
// 先绘制黑色背景提升可读性
cv::putText(output_image, plate_num, text_org,
cv::Scalar(0, 0, 0), // 颜色
myfont, // 字体对象
font_scale, // 字体尺寸
thickness + 2, // 线宽
cv::PutTextFlags::PUT_TEXT_ALIGN_LEFT, // 对齐方式
cv::Range(0, 0)); // 自动换行范围（0表示不换行）
cv::putText(output_image, plate_num, text_org, cv::Scalar(127, 0, 127), myfont, 10);
cout << "车牌 " << i+1 << ":\n"
<< " 位置: [" << roi.x << ", " << roi.y
<< ", " << roi.width << "x" << roi.height << "]\n"
<< " 置信度: " << det.score << "\n"
<< " 识别结果: " << plate_num << "\n" << endl;
cv::imshow("Detection Result", output_image);
}
}
cv::waitKey(0);
}
// 实时摄像头模式
else {
// 初始化设备连接
lockzhiner_vision_module::edit::Edit edit;
if (!edit.StartAndAcceptConnection()) {
std::cerr << "Error: Failed to start and accept connection." << std::endl;
return EXIT_FAILURE;
}
std::cout << "Device connected successfully." << std::endl;
cv::VideoCapture cap;
cap.set(cv::CAP_PROP_FRAME_WIDTH, 640);
cap.set(cv::CAP_PROP_FRAME_HEIGHT, 480);
cap.open(0);
if (!cap.isOpened()) {
cerr << "Error: Could not open camera" << endl;
return 1;
}
cout << "实时识别模式启动 (按ESC退出)..." << endl;
cv::Mat frame;
while (true) {
cap >> frame;
if (frame.empty()) continue;
auto results = detector.Predict(frame);
cv::Mat display_frame = frame.clone();
for (const auto& det : results) {
cv::Rect rect(
static_cast<int>(det.box.x),
static_cast<int>(det.box.y),
static_cast<int>(det.box.width),
static_cast<int>(det.box.height)
);
cv::rectangle(
display_frame,
rect,
cv::Scalar(0, 255, 0),
1,
cv::LINE_AA
);
}
// 添加时间戳
auto now = system_clock::now();
time_t now_time = system_clock::to_time_t(now);
cout << "\n===== " << ctime(&now_time)
<< "检测到 " << results.size() << " 个车牌 =====" << endl;
for (const auto& det : results) {
cv::Rect roi(det.box.x, det.box.y, det.box.width, det.box.height);
roi &= cv::Rect(0, 0, frame.cols, frame.rows);
if (roi.area() > 0) {
cv::Mat plate_img = frame(roi);
string plate_num = RecognizePlate(plate_img);
// 左上角偏移
cv::Point text_org(roi.x + 2, roi.y - 2);
// 先绘制黑色背景提升可读性
cv::putText(display_frame, plate_num, text_org,
cv::Scalar(0, 0, 0), // 颜色
myfont, // 字体对象
font_scale, // 字体尺寸
thickness + 2, // 线宽
cv::PutTextFlags::PUT_TEXT_ALIGN_LEFT, // 对齐方式
cv::Range(0, 0)); // 自动换行范围（0表示不换行）
cv::putText(display_frame, plate_num, text_org, cv::Scalar(127, 0, 127), myfont, 10);
cout << "[实时结果] 位置(" << roi.x << "," << roi.y
<< ") 识别: " << plate_num
<< " (置信度: " << det.score << ")" << endl;
}
}
edit.Print(display_frame);
// 退出检测
if (cv::waitKey(1) == 27) break;
}
}
return 0;
}
cmake_minimum_required(VERSION 3.10)
project(plate_recognize)
set(CMAKE_CXX_STANDARD 17)
set(CMAKE_CXX_STANDARD_REQUIRED ON)
# 定义项目根目录路径
set(PROJECT_ROOT_PATH "${CMAKE_CURRENT_SOURCE_DIR}/../..")
message("PROJECT_ROOT_PATH = " ${PROJECT_ROOT_PATH})
include("${PROJECT_ROOT_PATH}/toolchains/arm-rockchip830-linux-uclibcgnueabihf.toolchain.cmake")
# 定义 OpenCV SDK 路径
set(OpenCV_ROOT_PATH "${PROJECT_ROOT_PATH}/third_party/opencv-mobile-4.10.0-lockzhiner-vision-module")
set(OpenCV_DIR "${OpenCV_ROOT_PATH}/lib/cmake/opencv4")
find_package(OpenCV REQUIRED)
set(OPENCV_LIBRARIES "${OpenCV_LIBS}")
# 定义 LockzhinerVisionModule SDK 路径
set(LockzhinerVisionModule_ROOT_PATH "${PROJECT_ROOT_PATH}/third_party/lockzhiner_vision_module_sdk")
set(LockzhinerVisionModule_DIR "${LockzhinerVisionModule_ROOT_PATH}/lib/cmake/lockzhiner_vision_module")
find_package(LockzhinerVisionModule REQUIRED)
# ncnn配置
set(NCNN_ROOT_DIR "${PROJECT_ROOT_PATH}/third_party/ncnn-20240820-lockzhiner-vision-module") # 确保third_party层级存在
message(STATUS "Checking ncnn headers in: ${NCNN_ROOT_DIR}/include/ncnn")
# 验证头文件存在
if(NOT EXISTS "${NCNN_ROOT_DIR}/include/ncnn/net.h")
message(FATAL_ERROR "ncnn headers not found. Confirm the directory contains ncnn: ${NCNN_ROOT_DIR}")
endif()
set(NCNN_INCLUDE_DIRS "${NCNN_ROOT_DIR}/include")
set(NCNN_LIBRARIES "${NCNN_ROOT_DIR}/lib/libncnn.a")
add_executable(Test-plate_recognize plate_recognize.cc)
target_include_directories(Test-plate_recognize PRIVATE ${LOCKZHINER_VISION_MODULE_INCLUDE_DIRS} ${NCNN_INCLUDE_DIRS})
target_link_libraries(Test-plate_recognize PRIVATE ${OPENCV_LIBRARIES} ${NCNN_LIBRARIES} ${LOCKZHINER_VISION_MODULE_LIBRARIES})
install(
TARGETS Test-plate_recognize
RUNTIME DESTINATION .
)
使用 Docker Destop 打开 LockzhinerVisionModule 容器并执行以下命令来编译项目
# 进入Demo所在目录
cd /LockzhinerVisionModuleWorkSpace/LockzhinerVisionModule/Cpp_example/D09_plate_recognize
# 创建编译目录
rm -rf build && mkdir build && cd build
# 配置交叉编译工具链
export TOOLCHAIN_ROOT_PATH="/LockzhinerVisionModuleWorkSpace/arm-rockchip830-linux-uclibcgnueabihf"
# 使用cmake配置项目
cmake ..
# 执行编译项目
make -j8 && make install
在执行完上述命令后，会在build目录下生成可执行文件。
chmode 777 Test-plate_recognize
# 对车牌图片进行识别
./Test-plate_recognize Plate_recognize.rknn lpr2d.param lpr2d.bin image_path
# 摄像头实时识别
./Test-plate_recognize Plate_recognize.rknn lpr2d.param lpr2d.bin
通过上述内容，我们成功实现了一个车牌识别系统，包括：
此处可能存在不合适展示的内容，页面不予展示。您可通过相关编辑功能自查并修改。
如您确认内容无涉及 不当用语 / 纯广告导流 / 暴力 / 低俗色情 / 侵权 / 盗版 / 虚假 / 无价值内容或违法国家有关法律法规的内容，可点击提交进行申诉，我们将尽快为您处理。
马建仓 AI 助手