├── ObjectDetectionWithTrack.cpp
└── README.md


/ObjectDetectionWithTrack.cpp:
--------------------------------------------------------------------------------
  1 | #include <stdio.h>
  2 | #include <fstream>
  3 | #include <sstream>
  4 | #include <iostream>
  5 | 
  6 | #include <opencv2/dnn/dnn.hpp>
  7 | #include <opencv2/imgproc/imgproc.hpp>
  8 | #include <opencv2/highgui/highgui.hpp>
  9 | #include <opencv2/tracking/tracker.hpp>
 10 | 
 11 | 
 12 | using namespace cv;
 13 | using namespace dnn;
 14 | using namespace std;
 15 | 
 16 | // Initialize the parameters
 17 | vector<string> trackerTypes = {"BOOSTING", "MIL", "KCF", "TLD", "MEDIANFLOW", "GOTURN", "MOSSE", "CSRT"};
 18 | float confThreshold = 0.5; // Confidence threshold
 19 | float nmsThreshold = 0.4;  // Non-maximum suppression threshold
 20 | int inpWidth = 416;  // Width of network's input image
 21 | int inpHeight = 416; // Height of network's input image
 22 | vector<string> classes;
 23 | 
 24 | // Draw the predicted bounding box
 25 | void drawPred(int classId, float conf, int left, int top, int right, int bottom, Mat& frame)
 26 | {
 27 |     //Draw a rectangle displaying the bounding box
 28 |     rectangle(frame, Point(left, top), Point(right, bottom), Scalar(255, 178, 50), 3);
 29 |     
 30 |     //Get the label for the class name and its confidence
 31 |     string label = format("%.2f", conf);
 32 |     if (!classes.empty())
 33 |     {
 34 |         CV_Assert(classId < (int)classes.size());
 35 |         label = classes[classId] + ":" + label;
 36 |     }
 37 |     
 38 |     //Display the label at the top of the bounding box
 39 |     int baseLine;
 40 |     Size labelSize = getTextSize(label, FONT_HERSHEY_SIMPLEX, 0.5, 1, &baseLine);
 41 |     top = max(top, labelSize.height);
 42 |     rectangle(frame, Point(left, top - round(1.5*labelSize.height)), Point(left + round(1.5*labelSize.width), top + baseLine), Scalar(255, 255, 255), FILLED);
 43 |     putText(frame, label, Point(left, top), FONT_HERSHEY_SIMPLEX, 0.75, Scalar(0,0,0),1);
 44 | }
 45 | 
 46 | vector<Rect> postprocess(Mat& frame, const vector<Mat>& outs)
 47 | {
 48 |     vector<int> classIds;
 49 |     vector<float> confidences;
 50 |     vector<Rect> boxes;
 51 |     
 52 |     for (size_t i = 0; i < outs.size(); ++i)
 53 |     {
 54 |         // Scan through all the bounding boxes output from the network and keep only the
 55 |         // ones with high confidence scores. Assign the box's class label as the class
 56 |         // with the highest score for the box.
 57 |         float* data = (float*)outs[i].data;
 58 |         for (int j = 0; j < outs[i].rows; ++j, data += outs[i].cols)
 59 |         {
 60 |             Mat scores = outs[i].row(j).colRange(5, outs[i].cols);
 61 |             Point classIdPoint;
 62 |             double confidence;
 63 |             // Get the value and location of the maximum score
 64 |             minMaxLoc(scores, 0, &confidence, 0, &classIdPoint);
 65 |             if (confidence > confThreshold)
 66 |             {
 67 |                 int centerX = (int)(data[0] * frame.cols);
 68 |                 int centerY = (int)(data[1] * frame.rows);
 69 |                 int width = (int)(data[2] * frame.cols);
 70 |                 int height = (int)(data[3] * frame.rows);
 71 |                 int left = centerX - width / 2;
 72 |                 int top = centerY - height / 2;
 73 |                 
 74 |                 classIds.push_back(classIdPoint.x);
 75 |                 confidences.push_back((float)confidence);
 76 |                 boxes.push_back(Rect(left, top, width, height));
 77 |             }
 78 |         }
 79 |     }
 80 |     
 81 |     // Perform non maximum suppression to eliminate redundant overlapping boxes with
 82 |     // lower confidences
 83 |     vector<int> indices;
 84 |     NMSBoxes(boxes, confidences, confThreshold, nmsThreshold, indices);
 85 |     for (size_t i = 0; i < indices.size(); ++i)
 86 |     {
 87 |         int idx = indices[i];
 88 |         Rect box = boxes[idx];
 89 |         drawPred(classIds[idx], confidences[idx], box.x, box.y,
 90 |                  box.x + box.width, box.y + box.height, frame);
 91 |     }
 92 |     return boxes;
 93 | }
 94 | // Get the names of the output layers
 95 | vector<String> getOutputsNames(const Net& net)
 96 | {
 97 |     static vector<String> names;
 98 |     if (names.empty())
 99 |     {
100 |         //Get the indices of the output layers, i.e. the layers with unconnected outputs
101 |         vector<int> outLayers = net.getUnconnectedOutLayers();
102 |         
103 |         //get the names of all the layers in the network
104 |         vector<String> layersNames = net.getLayerNames();
105 |         
106 |         // Get the names of the output layers in names
107 |         names.resize(outLayers.size());
108 |         for (size_t i = 0; i < outLayers.size(); ++i)
109 |             names[i] = layersNames[outLayers[i] - 1];
110 |     }
111 |     return names;
112 | }
113 | 
114 | //用随机的颜色来画这些跟踪的框
115 | void getRandomColors(vector<Scalar> &color, int numColors)
116 | {
117 |     RNG rng(0);
118 |     for(int i = 0; i<numColors;i++)
119 |     {
120 |         color.push_back(Scalar(rng.uniform(0,255),rng.uniform(0, 255),rng.uniform(0, 255)));
121 |     }
122 |     
123 | }
124 | Ptr<Tracker> createTrackerByName(string trackerType)
125 | {
126 |     Ptr<Tracker> tracker;
127 |     if (trackerType ==  trackerTypes[0])
128 |         tracker = TrackerBoosting::create();
129 |     else if (trackerType == trackerTypes[1])
130 |         tracker = TrackerMIL::create();
131 |     else if (trackerType == trackerTypes[2])
132 |         tracker = TrackerKCF::create();
133 |     else if (trackerType == trackerTypes[3])
134 |         tracker = TrackerTLD::create();
135 |     else if (trackerType == trackerTypes[4])
136 |         tracker = TrackerMedianFlow::create();
137 |     else if (trackerType == trackerTypes[5])
138 |         tracker = TrackerGOTURN::create();
139 |     else if (trackerType == trackerTypes[6])
140 |         tracker = TrackerMOSSE::create();
141 |     else if (trackerType == trackerTypes[7])
142 |         tracker = TrackerCSRT::create();
143 |     else {
144 |         cout << "Incorrect tracker name" << endl;
145 |         cout << "Available trackers are: " << endl;
146 |         for (vector<string>::iterator it = trackerTypes.begin() ; it != trackerTypes.end(); ++it)
147 |             std::cout << " " << *it << endl;
148 |     }
149 |     return tracker;
150 | }
151 | int main()
152 | {
153 |     // Load names of classes
154 |     string classesFile = "/Users/tangxi/cpp/learnOpenCV/utils/coco.names";
155 |     ifstream ifs(classesFile.c_str());
156 |     string line;
157 |     while (getline(ifs, line)) classes.push_back(line);
158 |     
159 |     // Give the configuration and weight files for the model
160 |     String modelConfiguration = "/Users/tangxi/cpp/learnOpenCV/utils/yolov3.cfg";
161 |     String modelWeights = "/Users/tangxi/cpp/learnOpenCV/utils/yolov3.weights";
162 |     
163 |     // Load the network
164 |     Net net = readNetFromDarknet(modelConfiguration, modelWeights);
165 |     net.setPreferableBackend(DNN_BACKEND_OPENCV);
166 |     net.setPreferableTarget(DNN_TARGET_CPU);
167 |     string filename = "/Users/tangxi/cpp/learnOpenCV/utils/run.mp4";
168 |     VideoCapture cap(filename);
169 |     if(!cap.isOpened())
170 |     {
171 |         cerr<<"cannot open ..."<<endl;
172 |     }
173 |     Mat frame,blob;
174 |     int i = 0;
175 |     string trackerType = "CSRT";
176 |     Ptr<MultiTracker> multiTracker = MultiTracker::create();
177 |     vector<Scalar>colors;
178 |     vector<Rect>bboxes;
179 |     while(true)
180 |     {
181 |         cap>>frame;
182 |         if (frame.empty()) {
183 |             cout << "Done processing !!!" << endl;
184 |             waitKey(3000);
185 |             break;
186 |         }
187 |         if(i % 5 == 0)
188 |         {
189 |             // Create a 4D blob from a frame.
190 |             blobFromImage(frame, blob, 1/255.0, Size(inpWidth, inpHeight), Scalar(0,0,0), true, false);
191 |             
192 |             //Sets the input to the network
193 |             net.setInput(blob);
194 |             
195 |             // Runs the forward pass to get output of the output layers
196 |             vector<Mat> outs;
197 |             net.forward(outs, getOutputsNames(net));
198 |             bboxes = postprocess(frame,outs);
199 |             if(bboxes.size()<1)
200 |                 continue ;
201 | 
202 |             getRandomColors(colors, bboxes.size());
203 |             //为每个目标添加跟踪器
204 |             for(int i = 0; i<bboxes.size();i++)
205 |             {
206 |                 multiTracker->add(createTrackerByName(trackerType),frame,Rect2d(bboxes[i]));
207 |             }
208 |         }
209 |         //跟踪
210 |         multiTracker->update(frame);
211 |         
212 |         for(size_t i = 0; i<multiTracker->getObjects().size();i++)
213 |         {
214 |             rectangle(frame, multiTracker->getObjects()[i],  colors[i],2,1);
215 |         }
216 |         
217 |         imshow("MultiTracker",frame);
218 |         if(waitKey(1)==27)
219 |             break;
220 |         i++;
221 |         
222 |     }
223 |     return 0;
224 | }
225 | 


--------------------------------------------------------------------------------
/README.md:
--------------------------------------------------------------------------------
 1 | # Object-detection-with-tracking
 2 | ## Introduction
 3 | 视频目标检测会比较不稳定，而跟踪会比较稳定。因此在一段视频中每隔一段时间进行一次目标检测，对检测到的目标进行跟踪，增加实时目标检测的稳定性
 4 | ## Dependency 
 5 | - OpenCV 4.1.0  
 6 | - C++11
 7 | - Mac OS High Sierra 10.13.6, Xcode 10.1
 8 | 
 9 | ## Continue...
10 | 后续会添加代码注释和文档,以及代码结构优化.
11 | 


--------------------------------------------------------------------------------