当前位置： 首页 > article >正文

在android studio上使用rknn模块下面的yolov8_pose模型

article 2025/2/28 11:04:24

我的第一想法就是直接把rk的demo当成so库封装来用,我直接在yolov8_pose的c代码下面添加yolov8_pose.cc与yolov8_pose.h用作封装,先上代码

yolov8_pose.cc

#include "yolov8_pose.h"


#include <stdint.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>



#include "yolov8-pose.h"
#include "image_utils.h"
#include "file_utils.h"
#include "image_drawing.h"
#include "postprocess.h"
int skeleton[38] ={16, 14, 14, 12, 17, 15, 15, 13, 12, 13, 6, 12, 7, 13, 6, 7, 6, 8, 
            7, 9, 8, 10, 9, 11, 2, 3, 1, 2, 1, 3, 2, 4, 3, 5, 4, 6, 5, 7}; 

int ret;
rknn_app_context_t rknn_app_ctx;
image_buffer_t src_image;

int yolov8_init() {
	memset(&rknn_app_ctx, 0, sizeof(rknn_app_context_t));
	
	init_post_process();
	
    ret = init_yolov8_pose_model("/data/user/0/com.rockchip.gpadc.yolodemo/cache/yolov8_pose.rknn", &rknn_app_ctx);
    if (ret != 0) {
        printf("yolov8_init fail! ret=%d\n", ret);
    }
    return ret;
}

char* run_yolov8_pose(unsigned char* pixeldata,int cound) {

    memset(&src_image, 0, sizeof(image_buffer_t));
    //ret = read_image("/data/model/test.jpg", &src_image);
	
	 // 默认图像为3通道
    int w = 1280, h = 720, c = 3;
    // printf("load image wxhxc=%dx%dx%d path=%s\n", w, h, c, path);
    int msize = w * h * c;

    // 设置图像数据
    (&src_image)->virt_addr = pixeldata;
    (&src_image)->width = w;
    (&src_image)->height = h;
	(&src_image)->format = IMAGE_FORMAT_RGB888;
	
  //  if (ret != 0)
   // {
   //     printf("read image fail! ret=%d image_path=%s\n", ret, "/data/model/test.jpg");
   // }

    object_detect_result_list od_results;

    ret = inference_yolov8_pose_model(&rknn_app_ctx, &src_image, &od_results);
    if (ret != 0)
    {
        printf("inference_yolov8_pose_model fail! ret=%d\n", ret);
    }

	 char buffers[1024]; // 用于存储格式化后的字符串
	 char *finalBuffer = NULL; // 用于存储最终的字符串
	 // 分配足够的内存来存储所有字符串
     finalBuffer = (char *)malloc(4084);
     if (finalBuffer == NULL) {
         perror("内存分配失败");
     }
	 
	// 初始化最终缓冲区
    finalBuffer[0] = '\0';
		
	// 格式化字符串
     snprintf(reinterpret_cast<char *const>(buffers), sizeof(buffers), "%3d\n", "");
       

    // 画框和概率
    char text[256];
    
    for (int i = 0; i < od_results.count; i++)
    {
		//做一个限制人数的功能
		if(od_results.count <= cound){
			object_detect_result *det_result = &(od_results.results[i]);
		    // 将信息格式化到临时缓冲区
			//snprintf(buffers, 1024, "%s (left:%d top:%d right:%d bottom:%d) 概率:%.3f length:%d\n",
			//			coco_cls_to_name(det_result->cls_id),
			//			det_result->box.left, det_result->box.top,
			//			det_result->box.right, det_result->box.bottom,
			//			det_result->prop,od_results.count);
			
			snprintf(buffers, 1024, "%s (left:%d top:%d right:%d bottom:%d) 概率:%.3f length:%d\n",
						"person-pose",
						det_result->box.left, det_result->box.top,
						det_result->box.right, det_result->box.bottom,
						det_result->prop,od_results.count);
			
			
			
	
			// 将格式化后的字符串附加到 finalBuffer
			strcat(finalBuffer, buffers);

    
			for (int j = 0; j < 38/2; ++j)
			{
				int x1 = (int)(det_result->keypoints[skeleton[2*j]-1][0]);
				int y1 = (int)(det_result->keypoints[skeleton[2*j]-1][1]);
				int x2 = (int)(det_result->keypoints[skeleton[2*j+1]-1][0]);
				int y2 = (int)(det_result->keypoints[skeleton[2*j+1]-1][1]);
		
				// 将线段坐标添加到 finalBuffer
				snprintf(buffers, 1024, "skeleton:(%d, %d,%d, %d)\n", x1, y1, x2, y2);
				strcat(finalBuffer, buffers);
			}
			
			for (int j = 0; j < 17; ++j)
			{
				int cx = (int)(det_result->keypoints[j][0]);
				int cy = (int)(det_result->keypoints[j][1]);
				// 将关键点坐标添加到 finalBuffer
				snprintf(buffers, 1024, "key:(%d, %d)\n",cx, cy);
				strcat(finalBuffer, buffers);
			}
		}
        
    }

   // write_image("/data/model/out.jpg", &src_image);
	
 //}
	

    return finalBuffer;
}

void yolov8_cleanup() {
    deinit_post_process();
    ret = release_yolov8_pose_model(&rknn_app_ctx);
    if (ret != 0)
    {
        printf("release_yolov5_model fail! ret=%d\n", ret);
    }

    if (src_image.virt_addr != NULL)
    {

        free(src_image.virt_addr);
    }

}

yolov8_pose.h

#ifndef YOLOV8_POSE_H
#define YOLOV8_POSE_H



#ifdef __cplusplus
extern "C" {
#endif

// 初始化模型
int yolov8_init();

// 执行推理
char*  run_yolov8_pose(unsigned char* pixeldata,int cound);

// 清理工作
void yolov8_cleanup();

#ifdef __cplusplus
}
#endif

#endif // YOLOV8_POSE_H

1.代码解析

  1. 我仔细研究了一下代码,我发现,rk的demo是读取一张640*640的图片输入模型,然后再输出一张640*640带上骨架等信息的图片

  2. 我的功能逻辑是实时输入摄像头帧数据,然后输出骨架信息,人相框信息等

  3. 我这里默认使用1280*720的像素, 我这里写死了rknn模型的地址,你们可以自行更改

  4. 既然我把功能代码封装好了,我们就需要把这些功能代码打包成库,所以我们需要更改一下配置文件

  5. 这里的CMakeLists.txt文件我没有去做删改,我只是在原基础上增加打包库文件,并且使用所有的接口,如下源码

cmake_minimum_required(VERSION 3.10)
project(rknn_yolov8_pose_demo)

set(rknpu_yolov8-pose_file rknpu2/yolov8-pose.cc)

# 添加第三方库和工具库
add_subdirectory(${CMAKE_CURRENT_SOURCE_DIR}/../../../3rdparty/ 3rdparty.out)
add_subdirectory(${CMAKE_CURRENT_SOURCE_DIR}/../../../utils/ utils.out)

set(CMAKE_INSTALL_RPATH "$ORIGIN/lib")

file(GLOB SRCS ${CMAKE_CURRENT_SOURCE_DIR}/*.cc)

# 定义共享库 yolov8_pose
add_library(yolov8_pose SHARED 
    yolov8_pose.cc 
    postprocess.cc
    ${rknpu_yolov8-pose_file}
)

# 将包含目录添加到 yolov8_pose 库
target_include_directories(yolov8_pose PRIVATE
    ${CMAKE_CURRENT_SOURCE_DIR}                    
    ${LIBRKNNRT_INCLUDES}                         
    ${CMAKE_CURRENT_SOURCE_DIR}/../../../3rdparty   
    ${CMAKE_CURRENT_SOURCE_DIR}/../../../utils/   
)

# 定义一个变量，包含公共库
set(COMMON_LIBS
    imageutils
    fileutils
    imagedrawing    
    ${LIBRKNNRT}
    dl
)

# 将共享库 yolov8_pose 链接到相关库
target_link_libraries(yolov8_pose ${COMMON_LIBS})

# 添加可执行文件
add_executable(${PROJECT_NAME}
    main.cc
    postprocess.cc
    ${rknpu_yolov8-pose_file}
)

# 将可执行文件链接到共享库和公共库
target_link_libraries(${PROJECT_NAME}
    yolov8_pose
    ${COMMON_LIBS}  # 也链接公共库
)

if (CMAKE_SYSTEM_NAME STREQUAL "Android")
    target_link_libraries(${PROJECT_NAME}
        log
    )
endif()

if (CMAKE_SYSTEM_NAME STREQUAL "Linux")
    set(THREADS_PREFER_PTHREAD_FLAG ON)
    find_package(Threads REQUIRED)
    target_link_libraries(${PROJECT_NAME} Threads::Threads)
endif()

target_include_directories(${PROJECT_NAME} PRIVATE
    ${CMAKE_CURRENT_SOURCE_DIR}
    ${LIBRKNNRT_INCLUDES}
)

# 安装目标
install(TARGETS ${PROJECT_NAME} DESTINATION .)
install(FILES ${CMAKE_CURRENT_SOURCE_DIR}/../model/bus.jpg DESTINATION ./model)
install(FILES ${CMAKE_CURRENT_SOURCE_DIR}/../model/yolov8_pose_labels_list.txt DESTINATION ./model)
file(GLOB RKNN_FILES "${CMAKE_CURRENT_SOURCE_DIR}/../model/*.rknn")
install(FILES ${RKNN_FILES} DESTINATION model)

# 安装共享库
install(TARGETS yolov8_pose DESTINATION lib)

编译:这里就能在install下面找到我们编译成功的库

     2.android studio使用,直接使用ndk开发

    先上代码:

    native-lib.cc

    
#include <jni.h>
#include <unistd.h>
#include <errno.h>
#include <string.h>
#include <pthread.h>
#include <sys/syscall.h>
#include <sched.h>


#include <stdint.h>
#include <stdio.h>
#include <stdlib.h>

#include "yolov8_pose.h"

/******************************************************yolov8***************************************************/
// 其他必要的引入
#include <android/log.h>

//#define LOGI(...) ((void)__android_log_print(ANDROID_LOG_INFO, "YOLOv8PoseJNI", __VA_ARGS__))
#define LOGI(...) __android_log_print(ANDROID_LOG_INFO, "YOLOv8PoseJNI", ##__VA_ARGS__);

extern "C" JNIEXPORT void JNICALL
Java_com_rockchip_gpadc_demo_yolo_InferenceWrapper_native_1inityolov8(JNIEnv *env, jobject obj) {
    int ret = yolov8_init();
    LOGI("Java_com_rockchip_gpadc_demo_yolo_InferenceWrapper_native_1inityolov8  yolov8_init: %d",
         ret);
}

// 推理
extern "C" JNIEXPORT void JNICALL
Java_com_rockchip_gpadc_demo_yolo_InferenceWrapper_native_1inference(JNIEnv *env, jobject obj,
                                                                     jobject call_back,
                                                                     jbyteArray imageData,
                                                                     jint width, jint height) {
    // 创建一个 JNI 全局引用
    jobject global_callback = (*env).NewGlobalRef(call_back);
    // 获取回调方法的方法ID
    jclass cls = (*env).GetObjectClass(call_back);
    jmethodID gCallBackMid = (*env).GetMethodID(cls, "onCall", "(Ljava/lang/String;)V");
    // 获取字节数组的指针
    jbyte *byteArray = (*env).GetByteArrayElements(imageData, NULL);
    if (byteArray == nullptr) {
        return; // 获取数组元素失败
    }
    //我这里限制五个人 前五个 后面的不管 数据量太大的话可能刷新不过来
    char *result = run_yolov8_pose(reinterpret_cast<unsigned char *>(byteArray),5);

    // 将 char* 转换为 Java 字符串
    jstring resultString = env->NewStringUTF(result);
    // 调用回调方法，将结果字符串传递回 Java 侧
    env->CallVoidMethod(global_callback, gCallBackMid, resultString);
    // 释放 JNI 创建的字符串
    env->DeleteLocalRef(resultString); // 删除局部引用的字符串
    // 释放字节数组的元素
    env->ReleaseByteArrayElements(imageData, byteArray, 0);
    // 释放全局引用
    env->DeleteGlobalRef(global_callback);
}

// 清理资源
extern "C" JNIEXPORT void JNICALL
Java_com_rockchip_gpadc_demo_yolo_InferenceWrapper_native_1releaseModel(JNIEnv *env, jobject obj) {
    yolov8_cleanup();
}

    CMakeLists.txt

    # For more information about using CMake with Android Studio, read the
# documentation: https://d.android.com/studio/projects/add-native-code.html

# Sets the minimum version of CMake required to build the native library.

cmake_minimum_required(VERSION 3.4.1)

# Creates and names a library, sets it as either STATIC
# or SHARED, and provides the relative paths to its source code.
# You can define multiple libraries, and CMake builds them for you.
# Gradle automatically packages shared libraries with your APK.

# include rga.
#include_directories(src/main/cpp/rga)

add_library( # Sets the name of the library.
             rknn4j

             # Sets the library as a shared library.
             SHARED

             # Provides a relative path to your source file(s).
             src/main/cpp/native-lib.cc
             src/main/cpp/yolov8_pose.h
             )

# Searches for a specified prebuilt library and stores the path as a
# variable. Because CMake includes system libraries in the search path by
# default, you only need to specify the name of the public NDK library
# you want to add. CMake verifies that the library exists before
# completing its build.

find_library( # Sets the name of the path variable.
              log-lib

              # Specifies the name of the NDK library that
              # you want CMake to locate.
              log )

# Specifies libraries CMake should link to your target library. You
# can link multiple libraries, such as libraries you define in this
# build script, prebuilt third-party libraries, or system libraries.

target_link_libraries( # Specifies the target library.
                       rknn4j
                       # Links the target library to the log library
                       # included in the NDK.
                       ${CMAKE_SOURCE_DIR}/src/main/jniLibs/${ANDROID_ABI}/librknnrt.so
                       ${CMAKE_SOURCE_DIR}/src/main/jniLibs/${ANDROID_ABI}/librga.so
                       ${CMAKE_SOURCE_DIR}/src/main/jniLibs/${ANDROID_ABI}/libyolov8_pose.so
                       ${log-lib} )

     这里我在native-lib.cc使用了实时回调,把模型推理数据实时的回调给app,app调用

    package com.rockchip.gpadc.demo.yolo;

import com.rockchip.gpadc.demo.callstr;
/**
 * Created by randall on 18-4-18.
 */

public class InferenceWrapper {
    private final String TAG = "rkyolo.InferenceWrapper";

    static {
        System.loadLibrary("rknn4j");
    }
    public InferenceWrapper() {

    }


      public void setmessage(byte[] bytes, final callstr mcallstr){
         // Log.e("TAG","setmessage bytes==="+bytes.length);
          native_inference(new INativeListener() {
              @Override
              public void onCall(String str) {
                //  Log.e("TAG","模型回调数据======"+str);
                  mcallstr.onCall(str);
              }
          },bytes,1280,720);
     }

    public int initModel(){
        native_inityolov8();
        return 0; 
    }


    public void deinit() {
        native_releaseModel();
    }


    // java回调接口 INativeListener.java
    interface INativeListener {
        void onCall(String str);
    }


    /**************************************************yolov8******************************************/

    private native void native_inityolov8();
    private native void native_inference(INativeListener mINativeListener,byte[] imageData, int width, int height);
    private native void native_releaseModel();
}

     这里我做了个回调接口,在对应的地方调用即可

    这里还有个说明,因为rknn的推理数据是rgb的数据,但是很多摄像头的数据是yuv-nv12或者nv21,这里输入数据的时候需要转化一下,这个网上随便搜,很多.

    • 具体的打开摄像头,解析数据等代码因为保密原因这里我不能贴出来
    • 其实步骤很简单,在oncreat里面先初始化调用 
      public int initModel(){
    native_inityolov8();
    return 0; 
}
    • 然后转换一下数据格式传入jni,就能实时得到数据了 
      native_inference(new INativeListener() {
    @Override
    public void onCall(String str) {
        Log.e("TAG","模型回调数据======"+str);
        //解析
    }
},bytes,1280,720);
    • 具体怎么解析你们可以自定义上面的格式,这里我是一个数据一行,你们可以更改上面的yolov8_pose.cc里面的finalBuffer数据,这个数据就是所有数据回调
    • 具体就这些了,有什么不懂的可以找我咨询,我正常上班时间都会看一下消息.


    http://www.kler.cn/a/564446.html

    相关文章：

  1. 嘉孚朗SDP系列伺服螺丝刀使用介绍
  2. C# 确保程序只有一个实例运行
  3. shell脚本的相关练习---＞分支结构----＞循环结构
  4. yolo11 training benchmark [i7-13700 vs nvidia 3090]
  5. Staruml软件的介绍安装uml类图的绘制流程
  6. 第8章：流式海啸：数据重构
  7. 8、HTTP/1.0和HTTP/1.1的区别【高频】
  8. information_schema.processlist 表详解
  9. Stale file handle
  10. TCP 缓冲区核心机制
  11. 为什么要进行软件测试？
  12. 前端或者后端通常用到数组使用方式
  13. 华为对流程认知的三个层次（重新认识流程）
  14. 执行git操作时报错：`remote: [session-b8xxxda3] Access denied ...`解决方案
  15. 深度分析：AheadComputing的RISC-V技术背景、战略定位与挑战
  16. VScode在windows10上使用clang-format
  17. 《Python实战进阶》专栏 No 5：GraphQL vs RESTful API 对比与实现
  18. Spark基础篇-Application、Job、Stage 和 Task
  19. StrokesPlus【电脑鼠标键盘手势软件】v0.5.8.0 中文绿色便携版
  20. Scrapy：隧道代理中移除 Proxy-Authorization 的原理解析