S3315_RK3588/NeoTracker/vot_test/VOT_SOT_Ground_RGB.cpp

// 单目标对地跟踪流程测试:将TLD从算法中剥离到外部，导致API调用形式调整
// 读取avi视频进行测试
#include "NeoArithStandardDll.h"
#include <iostream>
#include <memory>
#include <string.h>
#include <algorithm>
#include <thread>
#include "opencv2/opencv.hpp"

#define TEST_WITH_AID 0     // 是否使用AI Detect
#define TEST_WITH_AIT 0    // 是否使用AI Tracker,如果设置为1，最外部的CMakeLists.txt 需要设置set(BUILD_AI_TRACK TRUE)

#define VOT_RECTANGLE
#include "vot.h"

#if TEST_WITH_AID
#include "Arith_YOLO_Detect.h"
#endif

#if TEST_WITH_AIT
#include "Arith_AITracker.h"
#endif

using std::cout;
using std::endl;

short SelectCX = 0;
short SelectCY = 0;
unsigned short  setLockBoxW = 0;
unsigned short  setLockBoxH = 0;

// 算法输入部分
ARIDLL_INPUTPARA stInputPara = { 0 };

// 算法输出部分
ARIDLL_OUTPUT stOutput = { 0 };

// AI Detect算法结果
#if TEST_WITH_AID
obj_res* g_pGLB_AIDetOutput = NULL;
int g_GLB_AIDetNum = 0;
#endif

// AI Tracker算法结果
#if TEST_WITH_AIT
API_AI_Tracker* g_GLB_AITracker = NULL;
AIT_OUTPUT g_GLB_AITrackOutput = { 0 };
#endif

#if TEST_WITH_AID
static void AIDetRun(ArithHandle pTracker, GD_VIDEO_FRAME_S img, int frameID)
{
    // 异步调用，考虑机器上传输延时，注意异步方式结果天然缓1帧。
    Async_YOLO_DetectTarget(img.u64VirAddr[0], img.u32Width, img.u32Height, frameID);
    g_pGLB_AIDetOutput = Async_YOLO_GetTargetArray(g_GLB_AIDetNum);

    int targetNum = 0;
    TARGET_OBJECT* pArray = ARIDLL_SearchFrameTargets(pTracker, img, &targetNum);

    int mergeNum = ARIDLL_MergeAITargets(pTracker, pArray, targetNum, g_pGLB_AIDetOutput, g_GLB_AIDetNum);
    stInputPara.nInputTargetNum = mergeNum;
    memcpy(stInputPara.stInputTarget, pArray, sizeof(TARGET_OBJECT) * mergeNum);

}
#endif



#if TEST_WITH_AIT
static int AITrackerRun(GD_VIDEO_FRAME_S img, int frameID)
{
    // 从传统算法输出中获取AI跟踪器的控制指令
    CENTERRECT32F InitBox = stOutput.stAI_TkCmd.InitBox;
    CENTERRECT32F TargetBox = stOutput.stAI_TkCmd.TargetBox;

    if (InitBox.w > 0 && InitBox.h > 0)
    {
        g_GLB_AITracker->init(img, InitBox);
        return 0;
    }
    if (!stOutput.stAI_TkCmd.bTrack)
    {
        g_GLB_AITracker->stopTrack();
        memset(&stInputPara.stAITrackerInfo, 0, sizeof(AIT_OUTPUT));
        memset(&g_GLB_AITrackOutput, 0, sizeof(AIT_OUTPUT));
        return 0;
    }

    g_GLB_AITracker->Track(img, TargetBox);

    // 获取跟踪结果
    g_GLB_AITracker->getTrackResult_Async(&g_GLB_AITrackOutput);

    // 向传统算法传参
    memcpy(&stInputPara.stAITrackerInfo, &g_GLB_AITrackOutput, sizeof(AIT_OUTPUT));

    return 0;
}
#endif

static void RunProcess(ArithHandle pTracker, GD_VIDEO_FRAME_S img)
{
#if TEST_WITH_AID
    // 运行AI识别算法
    AIDetRun(pTracker, img, stInputPara.unFrmId);
#endif

#if TEST_WITH_AIT
    // 运行SiamRPN跟踪算法
    AITrackerRun(img, stInputPara.unFrmId);
#endif
    // 调用TLD流程
    //ARIDLL_RunTLDTracker(pTracker, img);

    // 运行算法主控逻辑API
    ARIDLL_RunController(pTracker, img, stInputPara, &stOutput);
}

int main()
{
    VOT vot;

    cv::Rect initialization;
    initialization << vot.region();

    cv::Mat frame = cv::imread(vot.frame());

    int nWidth = frame.cols;
    int nHeight = frame.rows;

    // 创建算法句柄
    ArithHandle pTracker = STD_CreatEOArithHandle();

#if TEST_WITH_AID
    // AI检测初始化
    //YOLO_Init();
    Async_YOLO_Init();
#endif

#if TEST_WITH_AIT
    // AI跟踪器初始化
    g_GLB_AITracker = API_AI_Tracker::Create(AITrackerType::DaSaimRPN);
    g_GLB_AITracker->loadModel();

    memset(&g_GLB_AITrackOutput, 0, sizeof(AIT_OUTPUT));
#endif
    // 初始化为凝视-对地模式
    ARIDLL_EOArithInitWithMode(pTracker,nWidth,nHeight,GD_PIXEL_FORMAT_E::GD_PIXEL_FORMAT_RGB_PACKED,
    GLB_SYS_MODE::GLB_SYS_STARE,GLB_SCEN_MODE::GLB_SCEN_GROUND);


    // 构建图像类型
    GD_VIDEO_FRAME_S img = { 0 };
    img.enPixelFormat = GD_PIXEL_FORMAT_E::GD_PIXEL_FORMAT_RGB_PACKED;
    img.u32Width = nWidth;
    img.u32Height = nHeight;
    img.u32Stride[0] = img.u32Width * 3;
    img.u64VirAddr[0] = frame.data;

    stInputPara.unFreq = 30;
    stInputPara.stAirCraftInfo.stAtt.fYaw = 0;
    stInputPara.stAirCraftInfo.stAtt.fRoll = 0;
    stInputPara.stAirCraftInfo.stAtt.fPitch = 0;
    stInputPara.stServoInfo.fServoAz = 0;
    stInputPara.stServoInfo.fServoPt = 0;
    stInputPara.stCameraInfo.fPixelSize = 15;
    stInputPara.stCameraInfo.nFocus = 600;
    stInputPara.unFrmId = 0;

    // 调用一次进行算法内部的初始化
    RunProcess(pTracker, img);

    // 下发锁定
    SelectCX = initialization.x + initialization.width / 2;
    SelectCY = initialization.y + initialization.height / 2;
    setLockBoxW = initialization.width;
    setLockBoxH = initialization.height;

    ARIDLL_OBJINFO obj = { 0 };
    obj = ARIDLL_LockTarget(pTracker, img, SelectCX, SelectCY, setLockBoxW, setLockBoxH);

#if TEST_WITH_AIT
    if (obj.nObjW > 0)
    {
        // 使用EOTracker的锁定决策初始化AI跟踪器
        CENTERRECT32F initBox = { obj.nX,obj.nY, obj.nObjW, obj.nObjH };
        g_GLB_AITracker->init(img, initBox);
        // 获取跟踪结果
        g_GLB_AITracker->getTrackResult_Async(&g_GLB_AITrackOutput);
    }
#endif

    // 调用一次跟踪流程，完成在锁定帧的跟踪运行
    RunProcess(pTracker, img);

    // 模拟算法执行流程
    while(!vot.end())
    {

        stInputPara.unFrmId++;

        string imagepath = vot.frame();
        if (imagepath.empty())
        {
            break;
        }
        frame = cv::imread(imagepath);

        // 构建图像类型
        img.enPixelFormat = GD_PIXEL_FORMAT_E::GD_PIXEL_FORMAT_RGB_PACKED;
        img.u32Width = nWidth;
        img.u32Height = nHeight;
        img.u32Stride[0] = img.u32Width * 3;
        img.u64VirAddr[0] = frame.data;

        RunProcess(pTracker, img);

        auto trackerOut = stOutput.stTrackers[0];
        cv::Rect outRect;
        outRect.width = (int)trackerOut.nObjW;
        outRect.height= (int)trackerOut.nObjH;
        outRect.x = (int)trackerOut.nX - outRect.width / 2;
        outRect.y = (int)trackerOut.nY - outRect.height / 2;

        vot.report(outRect, trackerOut.fConf);
    }

    return 0;
}