1、优化识别切分效果，增加报警信息

2、来车检测后，对来车状态改为存储在txt中，模型推理时读取来判定帧所对应的车厢状态

config.yaml

@@ -114,7 +114,7 @@ model:
 
 gc_http_open: 1
 username: "guest_01"
-password: "d55b0f642e817eea24725d2f2a31dd08" # 神东
+password: "0a1688bce402a5f6a4279abbfc89c203" # 神东 d55b0f642e817eea24725d2f2a31dd08
 gc_http_url: "http://192.168.2.211:20004/api/train-carriage/identification/video-save"
 gc_gettoken_url: "http://192.168.2.211:20004/api/blade-auth/oauth/token"
 gc_image_srv: "http://192.168.2.211:9010/"
@@ -182,5 +182,5 @@ gc_c_space_frame_width: 500
 # 是否识别车头
 gc_train_heard_detect: true
 
-#过期文件夹天数
+# 识别结果保存天数
 gc_days_for_result_expire_folder: 3

nvidia_ascend_engine/common/AppCommon.h

@@ -516,6 +516,7 @@ typedef struct
     uint64_t i64TimeStamp = 0;              //帧数据时间戳
     std::shared_ptr<DecodedData> pDecodeData = nullptr;
     int iDirection = 0;                     //行驶方向(0-未知; 1-向左; 2-向右)
+    int nMonitorState = MONITOR_MODEL_INIT_STATE;
 } SaveImgData;
 
 //识别处理数据
@@ -895,6 +896,7 @@ typedef struct
     int iBigClassId = -1;                   //大框id (给第二步用 1-属性 23456-编号)
     uint32_t iFrameId = 0;                  //帧号
     float fCenterX; //第一步识别目标中心点X坐标
+    float fTargetWidth; // 第一步识别目标X坐标宽度
 } CalculateInfo;
 
 

nvidia_ascend_engine/common_engine/DataDealEngine/DataDealEngine.h

@@ -50,6 +50,7 @@ private:
     std::map<int, std::vector<std::string>> mapSourcePushPort_;
 
     MoveData moveData_;
+    std::queue<MoveData> queuwMoveData_;
     std::string strDataDir_;
 
     uint32_t iOrigDataNO_;      //原过车数据个数

nvidia_ascend_engine/common_engine/DataSourceEngine/CameraEngine.h

@@ -65,6 +65,8 @@ private:
     int nDelayTime;
     bool bIsAvc_ = false;
     int iAudioStream_ = -1;
+
+    bool bCameraError_ = false;
 };
 
 ENGINE_REGIST(CameraEngine)

nvidia_ascend_engine/common_engine/DeleteExpiredFolderEngine/DeleteExpiredFolderEngine.h

@@ -48,6 +48,7 @@ private:
 
 private:
     std::string strResultPath_;
+    std::string strLogBakPath_;
 
     int iDaysNumber_;
 };

nvidia_ascend_engine/common_engine/MergerEngine/MergerAllEngine.cpp

@@ -2,6 +2,22 @@
 
 using namespace ai_matrix;
 
+namespace
+{
+    std::map<std::string, std::string> mapLoad = {
+            {"C62", "60"},
+            {"C62B", "61"},
+            {"C63", "61"},
+            {"C64", "61"},
+            {"C65", "60"},
+            {"C70", "70"},
+            {"C80", "80"},
+            {"KM70", "70"},
+            {"KM81", "81"},
+            {"KM98", "98"}
+    };
+}
+
 MergerAllEngine::MergerAllEngine() {}
 
 MergerAllEngine::~MergerAllEngine() {}
@@ -113,6 +129,8 @@ void MergerAllEngine::PushData(std::shared_ptr<Train> pTrain)
             << "集装箱2: " << pTrain->container2.strContainerNo << "\n"
             << "集装箱2图片: " << pTrain->container2.strBestImg << "\n"
             << "集装箱2时间戳: " << pTrain->container2.i64TimeStamp << "\n"
+            << "车厢开始时间: " << MyUtils::getins()->Stamp2Time(pTrain->i64StartTimeStamp, true) << "\n"
+            << "车厢结束时间: " << MyUtils::getins()->Stamp2Time(pTrain->i64EndTimeStamp, true) << "\n"
             << " ---所有信息合并结果 END--- ";
     if (pTrain->bIsEnd)
     {
@@ -310,12 +328,28 @@ APP_ERROR MergerAllEngine::Process()
                 特殊处理：针对车头在后的且当前为最后一节，且未识别到车型，则默认设置为车头。
                 防止车尾间隔误识别，多切分车型，且web端无法过滤的问题。
                 */
-                if (!bHeadFrontFlag_ && pTrain->bIsEnd && pTrain->trainNum.iTrainTypeId == -1)
+                if (!bHeadFrontFlag_
+                    && pTrain->bIsEnd
+                    && pTrain->trainNum.iTrainTypeId == -1)
                 {
                     LogDebug << "cometime:" << pTrain->strTrainDate << " " << pTrain->strTrainName << " iCarXH:" << pTrain->iCarXH
                              << " num:" << pTrain->trainNum.strTrainNum;
                     pTrain->trainNum.iTrainTypeId = 0;
                 }
+
+                // 修正载重
+                for (std::map<std::string, std::string>::iterator it =
+                        mapLoad.begin();
+                     it != mapLoad.end(); ++it) {
+
+                    if (pTrain->trainNum.strTrainType.find(it->first) !=
+                        std::string::npos) {
+                        pTrain->trainPro.strLoad = it->second;
+                    }
+                }
+
+                // if (pTrain->trainPro.strLoad.size() > 2 && pTrain->trainPro.strLoad[0] != '1') pTrain->trainPro.strLoad = "";
+
                 PushData(pTrain);
 
                 //最后一节处理后，初始化参数

nvidia_ascend_engine/common_engine/SaveEngine/SaveImgEngine.cpp

@@ -154,6 +154,7 @@ APP_ERROR SaveImgEngine::Process()
             Json::Value jvFrameInfo;
             jvFrameInfo["timeStamp"] = pSaveImgData->i64TimeStamp;
             jvFrameInfo["status"] = iStatus;
+            jvFrameInfo["moveType"] = pSaveImgData->nMonitorState;
             jvFrameInfo["direction"] = pSaveImgData->iDirection;
             jvFrameInfo["width"] = iWidth;
             jvFrameInfo["height"] = iHeight;

nvidia_ascend_engine/common_engine/SocketEngine/SocketEngine.cpp

@@ -1,4 +1,4 @@
-#include "SocketEngine.h"
+#include "SocketEngine.h"
 
 
 
@@ -313,7 +313,7 @@ void* SocketEngine::heart_handler(void* arg)
                 ++it;
             }
         }
-		sleep(3);   // 定时三秒
+        sleep(1);   // 定时三秒
     }
 }
 

nvidia_ascend_engine/nvidia_engine/MoveEngine/MoveEngine.cpp

@@ -192,6 +192,8 @@ void MoveEngine::SingleDeviceProcess(std::shared_ptr<ProcessData> pProcessData,
         {
             outputQueMap_[engineName_ + "_" + std::to_string(engineId_) + "_" + std::to_string(*iter)]->push(std::static_pointer_cast<void>(pMoveData));
         }
+
+        // LogInfo << "【来车检测】" << pMoveData->iFrameId << "  isEnd:" << pProcessData->bIsEnd;
         //通知第一步开始识别        
         outputQueMap_[engineName_ + "_" + std::to_string(engineId_) + "_5"]->push(std::static_pointer_cast<void>(pMoveData));
     }
@@ -211,6 +213,7 @@ void MoveEngine::SingleDeviceProcess(std::shared_ptr<ProcessData> pProcessData,
     pSaveImgData->bIsEnd = pProcessData->bIsEnd;
     pSaveImgData->bSaveToFtp = true;
     pSaveImgData->i64TimeStamp = pProcessData->i64TimeStamp;
+    pSaveImgData->nMonitorState = nType;
     outputQueMap_[strPort0_]->push(std::static_pointer_cast<void>(pSaveImgData));
 }
 
@@ -282,16 +285,16 @@ APP_ERROR MoveEngine::Process()
 //            LogInfo<<"来车当前状态:"<< (nType == 0 ? "有车头" : (nType == 1 ? "无车"));
             switch (nType) {
                 case 0:
-                    LogDebug << "来车状态：有车头";
+                    LogDebug << iMoveDataNO_ * dataSourceCfg_.iSkipInterval << "来车状态：有车头";
                     break;
                 case 1:
-                    LogDebug << "来车状态：无车";
+                    LogDebug << iMoveDataNO_ * dataSourceCfg_.iSkipInterval << "来车状态：无车";
                     break;
                 case 2:
-                    LogDebug << "来车状态：车尾";
+                    LogDebug << iMoveDataNO_ * dataSourceCfg_.iSkipInterval << "来车状态：车尾";
                     break;
                 case 3:
-                    LogDebug << "来车状态：有车";
+                    LogDebug << iMoveDataNO_ * dataSourceCfg_.iSkipInterval << "来车状态：有车";
                     break;
             }
 

nvidia_ascend_engine/nvidia_engine/TrainStepOneEngine/TrainStepOneEngine.cpp

@@ -194,9 +194,9 @@ void TrainStepOneEngine::FilterInvalidInfo(std::vector<stDetection> &vecRet, std
     std::vector<stDetection> vecSpaceInfo;
     for (auto it = vecRet.begin(); it != vecRet.end();)
     {
-        LogDebug << "frameId:" << pProcessData->iFrameId
+        // LogDebug << "frameId:" << pProcessData->iFrameId
-                 << " bigclassid:" << it->class_id << " ltx:" << it->bbox[0] << " lty:" << it->bbox[1]
+        //          << " bigclassid:" << it->class_id << " ltx:" << it->bbox[0] << " lty:" << it->bbox[1]
-                 << " rbx:" << it->bbox[2] << " rby:" << it->bbox[3];
+        //          << " rbx:" << it->bbox[2] << " rby:" << it->bbox[3];
         // 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] >= dataSourceCfg.fIdentifyAreasLTX &&
               it->bbox[1] >= dataSourceCfg.fIdentifyAreasLTY &&
@@ -322,6 +322,22 @@ void TrainStepOneEngine::FilterInvalidInfo(std::vector<stDetection> &vecRet, std
             continue;
         }
 
+        if (it->class_id == K_TRAIN_NUM)
+        {
+            int iCenterY = pProcessData->iHeight / 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:"
+                        << pProcessData->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 = vecRet.erase(it);
+                continue;
+            }
+        }
+
         //补连塔的相机比较近，间隔基本在画面底部，因此当间隔比较靠画面上时过滤掉。
         if ((it->class_id >= 9 && it->class_id <= 17 && it->class_id != 15) || it->class_id == U_TRAIN_SPACE)
         {
@@ -484,10 +500,10 @@ APP_ERROR TrainStepOneEngine::Process()
                     SetTargetType(postSubData);
                     pPostData->vecPostSubData.emplace_back(postSubData);
 
-//                    LogDebug << "数据源:" << pProcessData->iDataSource << " 帧:" << pProcessData->iFrameId
+                    LogDebug << "数据源:" << pProcessData->iDataSource << " 帧:" << pProcessData->iFrameId
-//                             << " --iClassId:" << singledata.iClassId << " iLine:" << singledata.iLine << " confidence=" << singledata.fScore
+                             << " --iClassId:" << singledata.iClassId << " iLine:" << singledata.iLine << " confidence=" << singledata.fScore
-//                             << " lx=" << singledata.fLTX << " ly=" << singledata.fLTY << " rx=" << singledata.fRBX << " ry=" << singledata.fRBY
+                             << " lx=" << singledata.fLTX << " ly=" << singledata.fLTY << " rx=" << singledata.fRBX << " ry=" << singledata.fRBY
-//                             << " clear:" << singledata.fClear;
+                             << " clear:" << singledata.fClear;
                 }
             }
         }

nvidia_ascend_tools/nvidia_tools/yolov5/src/inference.cu

@@ -113,8 +113,8 @@ void Inference::doInference(IExecutionContext& context, cudaStream_t& stream, vo
 
 void Inference::doInferenceV2(IExecutionContext& context, cudaStream_t& stream, void **buffers, unsigned int outputIndex, float* output, int outputSize, int batchSize) 
 {
-    context.enqueue(batchSize, buffers, stream, nullptr);
+//    context.enqueue(batchSize, buffers, stream, nullptr);
-    //context.enqueueV2(buffers, stream, nullptr);  
+    context.enqueueV2(buffers, stream, nullptr);  
     CUDA_CHECK(cudaMemcpyAsync(output, buffers[outputIndex], batchSize * outputSize * sizeof(float), cudaMemcpyDeviceToHost, stream));  
     cudaStreamSynchronize(stream); 
 }