RV1126 NO.57:ROCKX+RV1126人脸识别推流项目之读取人脸图片并把特征值保存到sqlite3数据库

一.本章节介绍

本章节将讲解如何使用rockx提取人脸图像特征值,并将其存储到sqlite3数据库中。在实际开发中,人脸特征值通常都会存入数据库,常见的选择包括sqlite3、MySQL等。(注:本项目不会深入讲解数据库知识,而是带大家完成基础的增删改查操作)。

二.rockx读取人脸特征值保存数据库大体框图

上图展示了将人脸特征和图片数据写入数据库的处理流程:

  1. 连接并读取人脸SQLite3数据库
  2. 初始化三个Rockx句柄:
    • 人脸检测句柄(face_det_handle)
    • 人脸识别句柄(face_recognize_handle)
    • 人脸关键点句柄(face_5landmarks_handle)
  3. 使用rockx_image_read读取人脸图片
  4. 通过FILE API获取图片长度和二进制数据
  5. 使用rockx_face_align进行人脸图片对齐
  6. 调用rockx_face_recognize提取人脸特征值
  7. 将人脸特征值和图片二进制数据存入SQLite3数据库

三.rockx读取人脸特征值保存数据库的代码

3.1.Connection_sqlite3DataBase连接数据库

cpp 复制代码
  printf("Start Connection sqlite3......................\n");
  Connection_sqlite3DataBase();
  printf("End Connection_sqlite3DataBase......................\n");

这里封装了一个了SQLITE3连接数据库的函数Connection_sqlite3Database,这个函数的实现如下

cpp 复制代码
int Connection_sqlite3DataBase()
{
  rc = sqlite3_open("/userdata/face.db", &db);
  if (rc != SQLITE_OK)
  {
    fprintf(stderr, "Can't open database: %s\n", sqlite3_errmsg(db));
    sqlite3_close(db);
    exit(1);
  }
  else
  {
    printf("You have opened a sqlite3 database named bind.db successfully!\nCongratulation! Have fun!\n");
  }
  return 0;
}

这个函数直接调用了sqlite3的api sqlite3_open来初始化人脸数据库,若返回值不等于SQLITE_OK则初始化数据库失败,否则就初始化成功。

3.2.创建三个rockx句柄

cpp 复制代码
 // 1. 创建人脸检测句柄
  ret = rockx_create(&face_det_handle, ROCKX_MODULE_FACE_DETECTION_V2, config, 0);
  if (ret != ROCKX_RET_SUCCESS)
  {
    printf("init rockx module ROCKX_MODULE_FACE_DETECTION error %d\n", ret);
    return -1;
  }

  // 2. 创建人脸5关键点句柄
  ret = rockx_create(&face_5landmarks_handle, ROCKX_MODULE_FACE_LANDMARK_5,
                     config, 0);
  if (ret != ROCKX_RET_SUCCESS)
  {
    printf("init rockx module ROCKX_MODULE_FACE_LANDMARK_68 error %d\n", ret);
    return -1;
  }

  // 3. 创建人脸特征提取句柄
  ret = rockx_create(&face_recognize_handle, ROCKX_MODULE_FACE_RECOGNIZE,
                     config, 0);
  if (ret != ROCKX_RET_SUCCESS)
  {
    printf("init rockx module ROCKX_MODULE_FACE_LANDMARK_68 error %d\n", ret);
    return -1;
  }

我们需要创建三个人脸处理句柄来实现完整功能:

  1. 人脸检测句柄(ROCKX_MODULE_FACE_DETECTION_V2)
  2. 人脸识别句柄(ROCKX_MODULE_FACE_RECOGNIZE)
  3. 人脸关键点检测句柄(ROCKX_MODULE_FACE_LANDMARK_5)

这些句柄都通过rockx_create函数进行初始化。

3.3.读取人脸图片

cpp 复制代码
  rockx_image_t input_image1;
  rockx_image_read(image_path, &input_image1, 1); //读取图片文件到rockx_image_t(第二个参数1表示按RGB格式读取)

上图是读取人脸的图片,这里使用的是rockx_image_read来读取人脸图片,其中input_image是需要从命令行输入的。

3.4.使用FILE读取图片的二进制数据和长度

cpp 复制代码
FILE *imageFile = fopen(image_path, "rb"); // 打开名为image.jpg的图片文件(二进制模式)
  if (!imageFile)
  {
    printf("无法打开图片文件.\n");
    return -1;
  }

  fseek(imageFile, 0L, SEEK_END);   // 定位到文件尾部
  long fileSize = ftell(imageFile); // 获取文件大小
  rewind(imageFile);                // 重新定位到文件起始位置
  printf("fileSize = %ld\n", fileSize);

  unsigned char *buffer = new unsigned char[fileSize];                          // 分配足够大小的缓冲区
  size_t bytesRead = fread(buffer, sizeof(unsigned char), fileSize, imageFile); // 从文件中读取图像数据到缓冲区

3.5.使用 run_face_recognize检测人脸数据并找到最精确的人脸

cpp 复制代码
int run_face_recognize(const char *name, rockx_image_t *in_image, rockx_face_feature_t *out_feature, unsigned char * buffer, int buffer_size)
{
  rockx_ret_t ret;

  /*************** FACE Detect ***************/
  // create rockx_face_array_t for store result
  rockx_object_array_t face_array;
  memset(&face_array, 0, sizeof(rockx_object_array_t));

  // detect face
  ret = rockx_face_detect(face_det_handle, in_image, &face_array, NULL);
  if (ret != ROCKX_RET_SUCCESS)
  {
    printf("rockx_face_detect error %d\n", ret);
    return -1;
  }

  // process result
  for (int i = 0; i < face_array.count; i++)
  {
    int left = face_array.object[i].box.left;
    int top = face_array.object[i].box.top;
    int right = face_array.object[i].box.right;
    int bottom = face_array.object[i].box.bottom;
    float score = face_array.object[i].score;
    printf("%d box=(%d %d %d %d) score=%f\n", i, left, top, right, bottom,score);
  }

  rockx_object_t *max_face = get_max_face(&face_array);
  if (max_face == NULL)
  {
    printf("error no face detected\n");
    return -1;
  }

  // Face Align
  rockx_image_t out_img;
  memset(&out_img, 0, sizeof(rockx_image_t));
   /*************** 2. 人脸对齐 ***************/
  ret = rockx_face_align(face_5landmarks_handle, in_image, &(max_face->box),NULL, &out_img);
  if (ret != ROCKX_RET_SUCCESS)
  {
    return -1;
  }

   /*************** 3. 人脸特征提取 ***************/
  rockx_face_recognize(face_recognize_handle, &out_img, out_feature);
   /*************** 4. 特征+图片数据入库 ***************/
  insert_face_data_toDataBase(name, out_feature->feature, FEATURE_SIZE, buffer, buffer_size);

  // Release Aligned Image
  rockx_image_release(&out_img);

  return 0;
}



run_face_recognize(name, &input_image1, &out_feature1, buffer, bytesRead);
cpp 复制代码
rockx_object_t *get_max_face(rockx_object_array_t *face_array)
{
  if (face_array->count == 0)
  {
    return NULL;
  }
  rockx_object_t *max_face = NULL;//存储人脸的最大指针
  int i;
  for (i = 0; i < face_array->count; i++)
  {
    rockx_object_t *cur_face = &(face_array->object[i]);
    if (max_face == NULL)
    {
      max_face = cur_face;
      continue;
    }
    int cur_face_box_area = (cur_face->box.right - cur_face->box.left) *
                            (cur_face->box.bottom - cur_face->box.top);
    int max_face_box_area = (max_face->box.right - max_face->box.left) *
                            (max_face->box.bottom - max_face->box.top);
    if (cur_face_box_area > max_face_box_area)
    {
      max_face = cur_face;
    }
  }
  printf("get_max_face %d\n", i - 1);

  return max_face;
}

get_max_face 方法通过遍历所有人脸数据实现功能:计算每个人脸区域面积(cur_face_box_area )并与当前最大面积(max_face_box_area)比较。当检测到更大的人脸区域时,会更新最大面积值,最终确定max_face为最精确的人脸数据。

3.6. 把人脸特征值和图片数据保存到sqlite3数据库

cpp 复制代码
void insert_face_data_toDataBase(const char *name, float feature[512], int featureSize, uint8_t *image_data, int image_length)
{
  printf("face_size = %d\n", image_length);
  
  // 1. 预编译SQL并检查返回值
  int ret = sqlite3_prepare(db, "insert into face_data_table(name,face_feature,feature_size,image_data,image_size) values (?,?,?,?,?);", -1, &stmt, NULL);
  if (ret != SQLITE_OK) {
    printf("sqlite3_prepare error: %s\n", sqlite3_errmsg(db));
    return;
  }

  // 2. 绑定参数(修正特征字节数)
  ret = sqlite3_bind_text(stmt, 1, name, strlen(name), SQLITE_STATIC); // SQLITE_STATIC 等价于NULL,表示不释放
  if (ret != SQLITE_OK) goto err;
  ret = sqlite3_bind_blob(stmt, 2, feature, featureSize * sizeof(float), SQLITE_STATIC);
  if (ret != SQLITE_OK) goto err;
  ret = sqlite3_bind_int(stmt, 3, featureSize);
  if (ret != SQLITE_OK) goto err;
  ret = sqlite3_bind_blob(stmt, 4, image_data, image_length, SQLITE_STATIC);
  if (ret != SQLITE_OK) goto err;
  ret = sqlite3_bind_int(stmt, 5, image_length);
  if (ret != SQLITE_OK) goto err;

  // 3. 打印调试
  printf("insert face feature: ");
  for (int i = 0; i < 50; i++) {
    printf("%f ", feature[i]);
  }
  printf("\n");

  // 4. 执行插入并检查返回值
  ret = sqlite3_step(stmt);
  if (ret != SQLITE_DONE) {
    printf("sqlite3_step error: %s\n", sqlite3_errmsg(db));
  }

err:
  // 5. 释放语句对象
  sqlite3_finalize(stmt);
}

3.7.测试结果

(图二)

最后一步是将人脸特征值和图片数据存入SQLite3数据库。我们使用以下SQL插入语句:

sql 复制代码
INSERT INTO face_data_table(name, face_feature, feature_size, image_data, image_size) 
VALUES (?, ?, ?, ?, ?);

其中face_data_table是存储人脸数据的数据表,包含以下字段:

  • name:人脸名称
  • face_feature:人脸特征值
  • feature_size:特征值长度
  • image_data:图片数据
  • image_size:图片长度

通过sqlite3_prepare函数将SQL语句编译为可执行字节码,支持查询、插入、更新和删除等操作。具体参数绑定如下:

  1. sqlite3_bind_text(stmt, 1, name, strlen(name), NULL):绑定名称
  2. sqlite3_bind_blob(stmt, 2, feature, featureSize, NULL):绑定人脸特征值
  3. sqlite3_bind_int(stmt, 3, featureSize):绑定特征值长度
  4. sqlite3_bind_blob(stmt, 4, image_data, image_length, NULL):绑定图片数据
  5. sqlite3_bind_int(stmt, 5, image_length):绑定图片长度

3.8.完整代码

cpp 复制代码
/*
 * ./sqlite3_operation_test name image_path
 */

#include "rockx.h"
#include "sqlite3_operation.h"

#define FEATURE_SIZE 512

rockx_handle_t face_det_handle;
rockx_handle_t face_5landmarks_handle;
rockx_handle_t face_recognize_handle;



void insert_face_data_toDataBase(const char *name, float feature[512], int featureSize, uint8_t *image_data, int image_length)
{
  printf("face_size = %d\n", image_length);
  
  // 1. 预编译SQL并检查返回值
  int ret = sqlite3_prepare(db, "insert into face_data_table(name,face_feature,feature_size,image_data,image_size) values (?,?,?,?,?);", -1, &stmt, NULL);
  if (ret != SQLITE_OK) {
    printf("sqlite3_prepare error: %s\n", sqlite3_errmsg(db));
    return;
  }

  // 2. 绑定参数(修正特征字节数)
  ret = sqlite3_bind_text(stmt, 1, name, strlen(name), SQLITE_STATIC); // SQLITE_STATIC 等价于NULL,表示不释放
  if (ret != SQLITE_OK) goto err;
  ret = sqlite3_bind_blob(stmt, 2, feature, featureSize * sizeof(float), SQLITE_STATIC);
  if (ret != SQLITE_OK) goto err;
  ret = sqlite3_bind_int(stmt, 3, featureSize);
  if (ret != SQLITE_OK) goto err;
  ret = sqlite3_bind_blob(stmt, 4, image_data, image_length, SQLITE_STATIC);
  if (ret != SQLITE_OK) goto err;
  ret = sqlite3_bind_int(stmt, 5, image_length);
  if (ret != SQLITE_OK) goto err;

  // 3. 打印调试
  printf("insert face feature: ");
  for (int i = 0; i < 50; i++) {
    printf("%f ", feature[i]);
  }
  printf("\n");

  // 4. 执行插入并检查返回值
  ret = sqlite3_step(stmt);
  if (ret != SQLITE_DONE) {
    printf("sqlite3_step error: %s\n", sqlite3_errmsg(db));
  }

err:
  // 5. 释放语句对象
  sqlite3_finalize(stmt);
}



int Connection_sqlite3DataBase()
{
  rc = sqlite3_open("/userdata/face.db", &db);
  if (rc != SQLITE_OK)
  {
    fprintf(stderr, "Can't open database: %s\n", sqlite3_errmsg(db));
    sqlite3_close(db);
    exit(1);
  }
  else
  {
    printf("You have opened a sqlite3 database named bind.db successfully!\nCongratulation! Have fun!\n");
  }
  return 0;
}




rockx_object_t *get_max_face(rockx_object_array_t *face_array)
{
  if (face_array->count == 0)
  {
    return NULL;
  }
  rockx_object_t *max_face = NULL;//存储人脸的最大指针
  int i;
  for (i = 0; i < face_array->count; i++)
  {
    rockx_object_t *cur_face = &(face_array->object[i]);
    if (max_face == NULL)
    {
      max_face = cur_face;
      continue;
    }
    int cur_face_box_area = (cur_face->box.right - cur_face->box.left) *
                            (cur_face->box.bottom - cur_face->box.top);
    int max_face_box_area = (max_face->box.right - max_face->box.left) *
                            (max_face->box.bottom - max_face->box.top);
    if (cur_face_box_area > max_face_box_area)
    {
      max_face = cur_face;
    }
  }
  printf("get_max_face %d\n", i - 1);

  return max_face;
}

int run_face_recognize(const char *name, rockx_image_t *in_image, rockx_face_feature_t *out_feature, unsigned char * buffer, int buffer_size)
{
  rockx_ret_t ret;

  /*************** FACE Detect ***************/
  // create rockx_face_array_t for store result
  rockx_object_array_t face_array;
  memset(&face_array, 0, sizeof(rockx_object_array_t));

  // detect face
  ret = rockx_face_detect(face_det_handle, in_image, &face_array, NULL);
  if (ret != ROCKX_RET_SUCCESS)
  {
    printf("rockx_face_detect error %d\n", ret);
    return -1;
  }

  // process result
  for (int i = 0; i < face_array.count; i++)
  {
    int left = face_array.object[i].box.left;
    int top = face_array.object[i].box.top;
    int right = face_array.object[i].box.right;
    int bottom = face_array.object[i].box.bottom;
    float score = face_array.object[i].score;
    printf("%d box=(%d %d %d %d) score=%f\n", i, left, top, right, bottom,score);
  }

  rockx_object_t *max_face = get_max_face(&face_array);
  if (max_face == NULL)
  {
    printf("error no face detected\n");
    return -1;
  }

  // Face Align
  rockx_image_t out_img;
  memset(&out_img, 0, sizeof(rockx_image_t));
   /*************** 2. 人脸对齐 ***************/
  ret = rockx_face_align(face_5landmarks_handle, in_image, &(max_face->box),NULL, &out_img);
  if (ret != ROCKX_RET_SUCCESS)
  {
    return -1;
  }

   /*************** 3. 人脸特征提取 ***************/
  rockx_face_recognize(face_recognize_handle, &out_img, out_feature);
   /*************** 4. 特征+图片数据入库 ***************/
  insert_face_data_toDataBase(name, out_feature->feature, FEATURE_SIZE, buffer, buffer_size);

  // Release Aligned Image
  rockx_image_release(&out_img);

  return 0;
}


int main(int argc, char *argv[])
{
  rockx_ret_t ret;

  printf("----------------- main init ---------------------\n");
#if 1
  if (argc != 3)
  {
    printf("Usage: Insert DataBase  ./sqlite3_operation_test name image_path\n");
    return -1;
  }

  printf("Start Connection sqlite3......................\n");
  Connection_sqlite3DataBase();
  printf("End Connection_sqlite3DataBase......................\n");
    // 创建rockx配置对象
  rockx_config_t *config = rockx_create_config();
  // 设置rockx模型数据路径(需提前放置模型文件)
  rockx_add_config(config, ROCKX_CONFIG_DATA_PATH, "/userdata/rockx_data/");

  /*************** Creat Handle ***************/
  // create a face detection handle
   // 1. 创建人脸检测句柄
  ret = rockx_create(&face_det_handle, ROCKX_MODULE_FACE_DETECTION_V2, config, 0);
  if (ret != ROCKX_RET_SUCCESS)
  {
    printf("init rockx module ROCKX_MODULE_FACE_DETECTION error %d\n", ret);
    return -1;
  }

  // 2. 创建人脸5关键点句柄
  ret = rockx_create(&face_5landmarks_handle, ROCKX_MODULE_FACE_LANDMARK_5,
                     config, 0);
  if (ret != ROCKX_RET_SUCCESS)
  {
    printf("init rockx module ROCKX_MODULE_FACE_LANDMARK_68 error %d\n", ret);
    return -1;
  }

  // 3. 创建人脸特征提取句柄
  ret = rockx_create(&face_recognize_handle, ROCKX_MODULE_FACE_RECOGNIZE,
                     config, 0);
  if (ret != ROCKX_RET_SUCCESS)
  {
    printf("init rockx module ROCKX_MODULE_FACE_LANDMARK_68 error %d\n", ret);
    return -1;
  }

  const char *name = argv[1];
  const char *image_path = argv[2];

  rockx_face_feature_t out_feature1;
  rockx_image_t input_image1;
  rockx_image_read(image_path, &input_image1, 1); //读取图片文件到rockx_image_t(第二个参数1表示按RGB格式读取)

  FILE *imageFile = fopen(image_path, "rb"); // 打开名为image.jpg的图片文件(二进制模式)
  if (!imageFile)
  {
    printf("无法打开图片文件.\n");
    return -1;
  }

  fseek(imageFile, 0L, SEEK_END);   // 定位到文件尾部
  long fileSize = ftell(imageFile); // 获取文件大小
  rewind(imageFile);                // 重新定位到文件起始位置
  printf("fileSize = %ld\n", fileSize);

  unsigned char *buffer = new unsigned char[fileSize];                          // 分配足够大小的缓冲区
  size_t bytesRead = fread(buffer, sizeof(unsigned char), fileSize, imageFile); // 从文件中读取图像数据到缓冲区
   /*************** 执行人脸处理+入库 ***************/
  run_face_recognize(name, &input_image1, &out_feature1, buffer, bytesRead);
#endif
  return 0;
}

四.在板子上查询数据库的人脸数据

4.1. 打开sqlite3数据库

使用sqlite3 face.db打开人脸数据库

4.2. 查询人脸的数据

通过SQL查询语句检索人脸识别数据,执行命令为:

sql 复制代码
SELECT * FROM face_data_table

查询结果如上方截图所示。

相关推荐
boy快快长大2 小时前
【MySQL】InnoDB记录存储结构
数据库·mysql
yaoxtao2 小时前
neo4j数据库的导入和导出
数据库
卓码软件测评2 小时前
CMA/CNAS软件测评机构:【Gatling数据库性能关联测试JDBC连接和SQL执行时间监控】
数据库·sql·测试工具·性能优化·测试用例
lixora2 小时前
备份指定oracle block 防止误操作
数据库
凯子坚持 c2 小时前
Redis 数据库的服务器部署与 MCP 智能化交互深度实践指南
服务器·数据库·redis
好游科技2 小时前
IM即时通讯系统:安全可控、功能全面的社交解决方案全解析
安全·音视频·webrtc·im即时通讯·私有化部署im即时通讯·社交app
DBA小马哥2 小时前
Oracle迁移金仓全攻略:工业IOT场景下的易用性与安全保障
数据库·物联网·安全·oracle
‘胶己人’2 小时前
redis分布式锁
数据库·redis·分布式
初恋叫萱萱3 小时前
基于CodeRider-Kilo和MySQL开发一款书店管理系统
数据库·mysql·产品运营