465 lines
18 KiB
C++
465 lines
18 KiB
C++
#include "TrainStep1InferenceEngine.h"
|
||
#include <opencv2/opencv.hpp>
|
||
//#include "myqueue.h"
|
||
|
||
using namespace ai_matrix;
|
||
|
||
TrainStep1InferenceEngine::TrainStep1InferenceEngine() {}
|
||
|
||
TrainStep1InferenceEngine::~TrainStep1InferenceEngine() {}
|
||
|
||
APP_ERROR TrainStep1InferenceEngine::Init()
|
||
{
|
||
strPort0_ = engineName_ + "_" + std::to_string(engineId_) + "_0";
|
||
this->modelConfig_ = Config::getins()->getModelByTrainStep1Config();
|
||
this->dataSourceConfig_ = (engineId_ == 0) ?
|
||
Config::getins()->getDataSourceConfig() : Config::getins()->getSecondaryDataSourceConfig();
|
||
this->identifyConfig_ = Config::getins()->getIdentifyConfig();
|
||
|
||
int iFolderExist = access(modelConfig_.strModelPath.c_str(), R_OK);
|
||
if (iFolderExist == -1)
|
||
{
|
||
LogError << "模型：" << modelConfig_.strModelPath << " 不存在!";
|
||
return false;
|
||
}
|
||
class_num = this->modelConfig_.vecClass.size();
|
||
score_threshold = this->modelConfig_.fScoreThreshold;
|
||
|
||
int ret = initModel();
|
||
if (ret != APP_ERR_OK)
|
||
{
|
||
LogError << "Failed to read model info, ret = " << ret;
|
||
return ret;
|
||
}
|
||
|
||
LogInfo << "Step1InferenceEngine Init ok";
|
||
return APP_ERR_OK;
|
||
}
|
||
|
||
APP_ERROR TrainStep1InferenceEngine::initModel()
|
||
{
|
||
modelinfo.yolov5ClearityModelParam.uiClassNum = class_num;
|
||
modelinfo.yolov5ClearityModelParam.uiClearNum = clear_num;
|
||
modelinfo.yolov5ClearityModelParam.uiDetSize = det_size;
|
||
modelinfo.yolov5ClearityModelParam.fScoreThreshold = score_threshold;
|
||
modelinfo.yolov5ClearityModelParam.fNmsThreshold = nms_threshold;
|
||
|
||
modelinfo.modelCommonInfo.uiModelWidth = model_width;
|
||
modelinfo.modelCommonInfo.uiModelHeight = model_height;
|
||
modelinfo.modelCommonInfo.uiInputSize = input_size;
|
||
modelinfo.modelCommonInfo.uiOutputSize = output_size;
|
||
modelinfo.modelCommonInfo.uiChannel = INPUT_CHANNEL;
|
||
modelinfo.modelCommonInfo.uiBatchSize = batch_size;
|
||
modelinfo.modelCommonInfo.strInputBlobName = INPUT_BLOB_NAME;
|
||
modelinfo.modelCommonInfo.strOutputBlobName = OUTPUT_BLOB_NAME;
|
||
|
||
string strModelName = "";
|
||
|
||
int nRet = yolov5model.YoloV5ClearityInferenceInit(&modelinfo,
|
||
strModelName,
|
||
this->modelConfig_.strModelPath);
|
||
if (nRet != 0)
|
||
{
|
||
LogError << "YoloV5ClassifyInferenceInit nRet:" << nRet;
|
||
return APP_ERR_COMM_READ_FAIL;
|
||
}
|
||
return APP_ERR_OK;
|
||
}
|
||
|
||
APP_ERROR TrainStep1InferenceEngine::DeInit()
|
||
{
|
||
yolov5model.YoloV5ClearityInferenceDeinit();
|
||
LogInfo << "Step1InferenceEngine DeInit ok";
|
||
return APP_ERR_OK;
|
||
}
|
||
|
||
/**
|
||
* 获取第1步得分最高框
|
||
* inParam : std::vector<stDetection> &vecResult 推理符合结果
|
||
* outParam: std::vector<stDetection> &vecResult 每个类别得分最高结果
|
||
* return : N/A
|
||
*/
|
||
void TrainStep1InferenceEngine::getMaxScoreResult(std::vector<stDetection> &vecResult)
|
||
{
|
||
if (vecResult.size() < 2)
|
||
{
|
||
return;
|
||
}
|
||
std::map<Target, std::vector<stDetection>> mapResult;
|
||
for (size_t i = 0; i < vecResult.size(); i++)
|
||
{
|
||
stDetection stDTemp = vecResult.at(i);
|
||
if (stDTemp.class_id == 0)
|
||
{
|
||
mapResult[HEAD].emplace_back(stDTemp);
|
||
}
|
||
else if (stDTemp.class_id == 1)
|
||
{
|
||
mapResult[PRO].emplace_back(stDTemp);
|
||
}
|
||
else if ((stDTemp.class_id >= 2 && stDTemp.class_id <= 6) || stDTemp.class_id == 8 || stDTemp.class_id == 15)
|
||
{
|
||
mapResult[NUM].emplace_back(stDTemp);
|
||
}
|
||
else if (stDTemp.class_id >= 9 && stDTemp.class_id <= 17 && stDTemp.class_id != 15)
|
||
{
|
||
mapResult[TRAINSPACE].emplace_back(stDTemp);
|
||
}
|
||
else if (stDTemp.class_id == 18)
|
||
{
|
||
mapResult[SPACE].emplace_back(stDTemp);
|
||
}
|
||
else if (stDTemp.class_id == 7)
|
||
{
|
||
mapResult[CONTAINER].emplace_back(stDTemp);
|
||
}
|
||
}
|
||
|
||
//清空之前的结果
|
||
vecResult.clear();
|
||
// 每个类别中，获取得分最高的框
|
||
for (auto iter = mapResult.begin(); iter != mapResult.end(); iter++)
|
||
{
|
||
int iMaxPos = -1;
|
||
for (size_t i = 0; i < iter->second.size(); i++)
|
||
{
|
||
if (iMaxPos == -1)
|
||
{
|
||
iMaxPos = i;
|
||
}
|
||
else if (iter->second.at(i).class_conf > iter->second.at(iMaxPos).class_conf)
|
||
{
|
||
iMaxPos = i;
|
||
}
|
||
}
|
||
|
||
if (iMaxPos >= 0)
|
||
{
|
||
vecResult.emplace_back(iter->second.at(iMaxPos));
|
||
}
|
||
}
|
||
}
|
||
|
||
/**
|
||
* 设置大框类型
|
||
* inParam : PostSubData &postSubData :推理结果
|
||
* outParam: PostSubData &postSubData :推理结果
|
||
* return : N/A
|
||
*/
|
||
void TrainStep1InferenceEngine::getTargetType(SingleData &singleData)
|
||
{
|
||
if (singleData.iClassId == TRAIN_HEAD)
|
||
{
|
||
singleData.iTargetType = HEAD;
|
||
}
|
||
else if (singleData.iClassId == TRAIN_PRO)
|
||
{
|
||
singleData.iTargetType = PRO;
|
||
}
|
||
else if ((singleData.iClassId >= 2 && singleData.iClassId <= 6) ||
|
||
singleData.iClassId == J_TRAIN_NUM ||
|
||
singleData.iClassId == W_TRAIN_NUM)
|
||
{
|
||
singleData.iTargetType = NUM;
|
||
}
|
||
else if (singleData.iClassId >= 9 && singleData.iClassId <= 17 && singleData.iClassId != 15)
|
||
{
|
||
singleData.iTargetType = TRAINSPACE;
|
||
}
|
||
else if (singleData.iClassId == U_TRAIN_SPACE)
|
||
{
|
||
singleData.iTargetType = SPACE;
|
||
}
|
||
else if (singleData.iClassId == CONTAINERNUM)
|
||
{
|
||
singleData.iTargetType = CONTAINER;
|
||
}
|
||
}
|
||
|
||
/**
|
||
* 过滤无效信息
|
||
* inParam : std::vector<stDetection> &vecRet :识别结果数据
|
||
: std::shared_ptr<ProcessData> pProcessData :帧信息数据
|
||
* outParam: N/A
|
||
* return : N/A
|
||
*/
|
||
void TrainStep1InferenceEngine::filterInvalidInfo(std::vector<stDetection> &vecInferenceResult,
|
||
std::shared_ptr<VTrainStep1Data> &pVTrainStep1Data)
|
||
{
|
||
std::vector<stDetection> vecSpaceInfo;
|
||
for (auto it = vecInferenceResult.begin(); it != vecInferenceResult.end();)
|
||
{
|
||
// LogDebug << " 帧:" << pVTrainStep1Data->iFrameId
|
||
// << " --iClassId:" << it->class_id
|
||
//// << " iLine:" << it->clear_conf
|
||
// << " confidence=" << it->class_conf
|
||
// << " lx=" << it->bbox[0]
|
||
// << " ly=" << it->bbox[1]
|
||
// << " rx=" << it->bbox[2]
|
||
// << " ry=" << it->bbox[3]
|
||
// << " clear:" << it->clear_conf;
|
||
// 根据配置文件中 设置的识别范围，过滤掉无效数据
|
||
if (!(it->bbox[0] >= this->dataSourceConfig_.vecIdentifyAreas[0] &&
|
||
it->bbox[1] >= this->dataSourceConfig_.vecIdentifyAreas[1] &&
|
||
it->bbox[2] <= this->dataSourceConfig_.vecIdentifyAreas[2] &&
|
||
it->bbox[3] <= this->dataSourceConfig_.vecIdentifyAreas[3]))
|
||
{
|
||
LogDebug << "frameId:" << pVTrainStep1Data->iFrameId
|
||
<< " 类别:" << it->class_id << " 超出识别区域-识别区域：("
|
||
<< this->dataSourceConfig_.vecIdentifyAreas[0] << ","
|
||
<< this->dataSourceConfig_.vecIdentifyAreas[1] << "),("
|
||
<< this->dataSourceConfig_.vecIdentifyAreas[2] << ","
|
||
<< this->dataSourceConfig_.vecIdentifyAreas[2] << ")";
|
||
it = vecInferenceResult.erase(it);
|
||
continue;
|
||
}
|
||
|
||
// 如果设置了不识别车头，则去掉车头标记的大框 !this->identifyConfig_.bTrainHeardDetect &&
|
||
if (it->class_id == TRAIN_HEAD)
|
||
{
|
||
LogDebug << "frameId:" << pVTrainStep1Data->iFrameId << " 过滤掉车头编号";
|
||
it = vecInferenceResult.erase(it);
|
||
continue;
|
||
}
|
||
|
||
// 去除车头时的非车头编号信息
|
||
if(pVTrainStep1Data->iTrainStage == MONITOR_MODEL_TRAIN_HEAD )
|
||
{
|
||
if(it->class_id != TRAIN_HEAD)
|
||
{
|
||
LogDebug << " 帧号:" << pVTrainStep1Data->iFrameId
|
||
<< " 大类:" << it->class_id << " 识别于车头位置，无效！";
|
||
it = vecInferenceResult.erase(it);
|
||
continue;
|
||
}
|
||
}
|
||
|
||
// 去除车尾的车头编号信息
|
||
if (pVTrainStep1Data->iTrainStage != MONITOR_MODEL_TRAIN_HEAD)
|
||
{
|
||
if (it->class_id == TRAIN_HEAD)
|
||
{
|
||
LogDebug << " 帧号:" << pVTrainStep1Data->iFrameId
|
||
<< " 大类:" << it->class_id << " 识别于非车头位置，无效！";
|
||
it = vecInferenceResult.erase(it);
|
||
continue;
|
||
}
|
||
}
|
||
|
||
// 去除车尾 和 车头车体之间 的间隔信息
|
||
if ((pVTrainStep1Data->iTrainStage == MONITOR_MODEL_TRAIN_TAIL || pVTrainStep1Data->iTrainStage == MONITOR_MODEL_HEAD_FIRST)
|
||
&& (it->class_id >= C_TRAIN_SPACE && it->class_id <= U_TRAIN_SPACE && it->class_id != W_TRAIN_NUM))
|
||
{
|
||
LogDebug << " frameId:" << pVTrainStep1Data->iFrameId
|
||
<< " bigclassid:" << it->class_id
|
||
<<" 识别于车尾或者车头与车身交接部分,无效！";
|
||
it = vecInferenceResult.erase(it);
|
||
continue;
|
||
}
|
||
|
||
// 过滤掉识别于模型反馈无车状态下的所有大框信息
|
||
if (pVTrainStep1Data->iTrainStage == MONITOR_MODEL_NO_TRAIN)
|
||
{
|
||
LogDebug << " frameId:" << pVTrainStep1Data->iFrameId
|
||
<< " bigclassid:" << it->class_id
|
||
<<" 识别于模型反馈的无车状态下,无效！";
|
||
it = vecInferenceResult.erase(it);
|
||
continue;
|
||
}
|
||
|
||
//剔除高度大于宽的车号大框
|
||
if (((it->class_id >= K_TRAIN_NUM && it->class_id <= NX_TRAIN_NUM)
|
||
|| it->class_id == J_TRAIN_NUM
|
||
|| it->class_id == W_TRAIN_NUM) &&
|
||
(it->bbox[3] - it->bbox[1]) > (it->bbox[2] - it->bbox[0]))
|
||
{
|
||
LogWarn << " frameId:" << pVTrainStep1Data->iFrameId
|
||
<< " bigclassid:" << it->class_id << " 过滤 高度大于宽度的车号";
|
||
it = vecInferenceResult.erase(it);
|
||
continue;
|
||
}
|
||
|
||
if (it->class_id == K_TRAIN_NUM)
|
||
{
|
||
int iCenterY = IMAGE_HEIGHT / 2;
|
||
int iHeight0 = it->bbox[1] / 2 + it->bbox[3] / 2;
|
||
if (iHeight0 > iCenterY) {
|
||
LogWarn << "矿车编号大框在画面Y轴中线以下，帧号:"
|
||
<< pVTrainStep1Data->iFrameId
|
||
<< " 画面Y轴中心：" << iCenterY
|
||
<< " 大框Y轴中心：" << iHeight0 ;
|
||
// << "[" << it->bbox[0] << "," << it->bbox[1] << "]"
|
||
// << "[" << it->bbox[2] << "," << it->bbox[3] << "]";
|
||
it = vecInferenceResult.erase(it);
|
||
continue;
|
||
}
|
||
}
|
||
|
||
if (it->class_id >= C_TRAIN_SPACE && it->class_id <= U_TRAIN_SPACE && it->class_id != W_TRAIN_NUM)
|
||
{
|
||
vecSpaceInfo.emplace_back(*it);
|
||
}
|
||
++it;
|
||
}
|
||
|
||
if (vecInferenceResult.size() <= 0) return;
|
||
|
||
// 过滤与间隔X轴重合的其他大框
|
||
// for (auto it = vecInferenceResult.begin(); it != vecInferenceResult.end();)
|
||
// {
|
||
// if (!((it->class_id >= 9 && it->class_id <= 17 && it->class_id != 15) || it->class_id == 18))
|
||
// {
|
||
// for (int i = 0; i < vecSpaceInfo.size(); i++)
|
||
// {
|
||
// if (
|
||
// (it->bbox[0] > vecSpaceInfo[i].bbox[0]
|
||
// && it->bbox[0] < vecSpaceInfo[i].bbox[2])
|
||
// ||
|
||
// (it->bbox[2] > vecSpaceInfo[i].bbox[0]
|
||
// && it->bbox[2] < vecSpaceInfo[i].bbox[2])
|
||
// )
|
||
// {
|
||
// LogWarn << "-- " << it->class_id << " _ " << vecSpaceInfo[i].class_id
|
||
// << " " << it->bbox[0] << "," << it->bbox[2] << " || " << vecSpaceInfo[i].bbox[0] << "," << vecSpaceInfo[i].bbox[2];
|
||
// it = vecInferenceResult.erase(it);
|
||
// break;
|
||
// }
|
||
// }
|
||
// }
|
||
// }
|
||
|
||
//主摄像头1帧如果只识别2个大框，如果非平车的车号和属性场景，则必有间隔框
|
||
if (vecInferenceResult.size() >= 2)
|
||
{
|
||
int iHeight0 = vecInferenceResult[0].bbox[1] / 2 + vecInferenceResult[0].bbox[3] / 2;
|
||
int iHeight1 = vecInferenceResult[1].bbox[1] / 2 + vecInferenceResult[1].bbox[3] / 2;
|
||
int iCenterY = IMAGE_HEIGHT / 2;
|
||
if (iHeight0 < iCenterY && iHeight1 < iCenterY) //非平车
|
||
{
|
||
bool bHaveSpace = false;
|
||
for (auto &it : vecInferenceResult)
|
||
{
|
||
if (it.class_id >= C_TRAIN_SPACE
|
||
&& it.class_id <= U_TRAIN_SPACE
|
||
&& it.class_id != W_TRAIN_NUM)
|
||
{
|
||
bHaveSpace = true;
|
||
}
|
||
}
|
||
|
||
if (!bHaveSpace)
|
||
{
|
||
LogDebug << " frameId:" << pVTrainStep1Data->iFrameId << " no space";
|
||
vecInferenceResult.clear();
|
||
}
|
||
// if (!(vecInferenceResult[0].class_id >= C_TRAIN_SPACE
|
||
// && vecInferenceResult[0].class_id <= U_TRAIN_SPACE
|
||
// && vecInferenceResult[0].class_id != W_TRAIN_NUM)
|
||
// && !(vecInferenceResult[1].class_id >= C_TRAIN_SPACE
|
||
// && vecInferenceResult[1].class_id <= U_TRAIN_SPACE
|
||
// && vecInferenceResult[1].class_id != W_TRAIN_NUM))
|
||
// {
|
||
// LogDebug << " frameId:" << pVTrainStep1Data->iFrameId << " no space";
|
||
// vecInferenceResult.clear();
|
||
// }
|
||
}
|
||
}
|
||
}
|
||
|
||
APP_ERROR TrainStep1InferenceEngine::Process()
|
||
{
|
||
int iRet = APP_ERR_OK;
|
||
|
||
while (!isStop_)
|
||
{
|
||
std::shared_ptr<void> pVoidData0 = nullptr;
|
||
inputQueMap_[strPort0_]->pop(pVoidData0);
|
||
if (nullptr == pVoidData0)
|
||
{
|
||
usleep(1000); //1ms
|
||
continue;
|
||
}
|
||
|
||
std::shared_ptr<VTrainStep1Data> pVTrainStep1Data = std::static_pointer_cast<VTrainStep1Data>(pVoidData0);
|
||
|
||
if (pVTrainStep1Data->cvImage.empty())
|
||
{
|
||
if (pVTrainStep1Data->bIsEnd)
|
||
{
|
||
std::shared_ptr<InferenceResultData> pInferenceResultData = std::make_shared<InferenceResultData>();
|
||
|
||
pInferenceResultData->iDataSource = pVTrainStep1Data->iDataSource;
|
||
pInferenceResultData->bIsEnd = pVTrainStep1Data->bIsEnd;
|
||
outputQueMap_[engineName_ + "_" + std::to_string(engineId_) + std::to_string(pInferenceResultData->iDataSource)]->push(std::static_pointer_cast<void>(pInferenceResultData), true);
|
||
}
|
||
usleep(1000); //1ms
|
||
continue;
|
||
}
|
||
// else
|
||
// {
|
||
// vector<int> compression_params;
|
||
// compression_params.push_back(cv::IMWRITE_JPEG_QUALITY); //选择jpeg
|
||
// compression_params.push_back(100); //图片质量
|
||
// cv::imwrite("./jpg/" + std::to_string(pVTrainStep1Data->iFrameId) + ".jpg", pVTrainStep1Data->cvImage, compression_params);
|
||
//
|
||
// }
|
||
|
||
//进行推理
|
||
std::vector<stDetection> vecInferenceResult;
|
||
yolov5model.YoloV5ClearityInferenceModel(pVTrainStep1Data->cvImage, vecInferenceResult);
|
||
|
||
//过滤无效信息
|
||
this->filterInvalidInfo(vecInferenceResult, pVTrainStep1Data);
|
||
|
||
this->getMaxScoreResult(vecInferenceResult);
|
||
|
||
std::shared_ptr<InferenceResultData> pInferenceResultData = std::make_shared<InferenceResultData>();
|
||
|
||
pInferenceResultData->iDataSource = pVTrainStep1Data->iDataSource;
|
||
pInferenceResultData->iFrameId = pVTrainStep1Data->iFrameId;
|
||
pInferenceResultData->bIsEnd = pVTrainStep1Data->bIsEnd;
|
||
pInferenceResultData->strTrainDate = pVTrainStep1Data->strTrainDate;
|
||
pInferenceResultData->strTrainTime = pVTrainStep1Data->strTrainTime;
|
||
pInferenceResultData->strTrainName = pVTrainStep1Data->strTrainName;
|
||
|
||
for (size_t j = 0; j < vecInferenceResult.size(); j++)
|
||
{
|
||
/*
|
||
[0:车头; 1:属性; 2:煤炭漏斗车(兖矿自备,枣矿自备); 3:敞车; 4:棚车; 5:罐车; 6:平车
|
||
7:集装箱; 8:牲畜车; 9:敞车间隔; 10:自备车间隔; 11:平车间隔; 12:罐车间隔; 13:棚车车间隔;
|
||
14:牲畜车间隔; 15:毒品车; 16: 毒品车间隔; 17:混合车厢间隔; 18:连接轴通用间隔; 19:集装箱号; 20:倒集装箱号]
|
||
*/
|
||
if (vecInferenceResult[j].class_id < 0 || vecInferenceResult[j].class_id > 20)
|
||
{
|
||
continue;
|
||
}
|
||
SingleData singledata;
|
||
singledata.iClassId = vecInferenceResult[j].class_id;
|
||
singledata.fScore = vecInferenceResult[j].class_conf;
|
||
singledata.fLTX = vecInferenceResult[j].bbox[0];
|
||
singledata.fLTY = vecInferenceResult[j].bbox[1];
|
||
singledata.fRBX = vecInferenceResult[j].bbox[2];
|
||
singledata.fRBY = vecInferenceResult[j].bbox[3];
|
||
singledata.fClear = vecInferenceResult[j].clear_id;
|
||
|
||
this->getTargetType(singledata);
|
||
pInferenceResultData->vecSingleData.emplace_back(singledata);
|
||
|
||
LogDebug << " 帧:" << pInferenceResultData->iFrameId
|
||
<< " 数据源:" << pInferenceResultData->iDataSource
|
||
<< " --iClassId:" << singledata.iClassId
|
||
<< " iLine:" << singledata.iLine
|
||
<< " confidence=" << singledata.fScore
|
||
<< " lx=" << singledata.fLTX
|
||
<< " ly=" << singledata.fLTY
|
||
<< " rx=" << singledata.fRBX
|
||
<< " ry=" << singledata.fRBY
|
||
<< " clear:" << singledata.fClear;
|
||
}
|
||
|
||
pVTrainStep1Data->cvImage.release();
|
||
|
||
outputQueMap_[this->strPort0_]->push(std::static_pointer_cast<void>(pInferenceResultData), true);
|
||
}
|
||
return APP_ERR_OK;
|
||
} |