ROS：kcf算法+行人检测 = 让机器人自动识别并追踪行人

Posted 2020-08-23 ImportSUC

实现目标：机器人检测到有人走过来，迎上去并开始追踪。
追踪算法使用kcf算法，关于kcf追踪的ros库在github地址 https://github.com/TianyeAlex/tracker_kcf_ros，kcf算法是目前追踪算法中比较好的，程序跑起来后效果也是不错的。我能力有限，在这里不作介绍。有兴趣的可以去研究一下。这里主要讲一下在次基础上添加行人检测，做到自动追踪。
训练库地址：http://download.csdn.net/detail/yiranhaiziqi/9711174，下载后放到src目录下。

追踪的代码结构
作者将kcf算法封装起来，在runtracker.cpp里面调用。

程序跑起来的效果

出现一个窗口，用鼠标左键选中一个区域作为感兴趣区域，之后机器人会跟踪这个区域。例如，选中画面中的椅子，移动椅子之后，机器人会跟随移动。选中画面中的人或者人的某个部位都可以实现人物跟踪。我要想实现自动追踪，就是把鼠标选择跟踪物变成自动选择跟踪物，这里的跟踪物就是行人。
首先要先实现行人检测，在opencv中，有行人检测的demo，路径在opencv-2.4.13/samples/cpp/peopledetect.cpp。接下来做的就是把代码结合起来。
简单介绍一下runtracker.cpp。
ImageConverter类是核心
初始化我们要接受/发送主题的Publisher 和Subscriber，设置相应的回掉函数。

image_sub_ = it_.subscribe("/camera/rgb/image_rect_color", 1,&ImageConverter::imageCb, this);
        depth_sub_ = it_.subscribe("/camera/depth/image", 1,&ImageConverter::depthCb, this);
        pub = nh_.advertise<geometry_msgs::Twist>("/mobile_base/mobile_base_controller/cmd_vel", 1000);

image_sub_是接受rgb图的subscribe，执行imageCb回掉函数，imageCb主要是将摄像头的数据显示在窗口中，选择感兴趣区域。
depth_sub_是接受深度图的subscribe，执行depthCb回掉函数，depthCb作用就是计算距离和方向。
了解到这里之后，要将手动选择感兴趣区域改为自动选择感兴趣区域，必然是在imageCb函数中修改。
imageCb中 cv::setMouseCallback(RGB_WINDOW, onMouse, 0);监听鼠标操作，如果鼠标不动，程序不会往下执行。onMouse为鼠标监听回调。要实现自动选择肯定就不能用这个了，将其注掉。
再来看下onMouse函数做了什么事

void onMouse(int event, int x, int y, int, void*)
{
    if (select_flag)
    {
        selectRect.x = MIN(origin.x, x);
        selectRect.y = MIN(origin.y, y);
        selectRect.width = abs(x - origin.x);
        selectRect.height = abs(y - origin.y);
        selectRect &= cv::Rect(0, 0, rgbimage.cols, rgbimage.rows);
    }
    if (event == CV_EVENT_LBUTTONDOWN)
    {
        bBeginKCF = false;
        select_flag = true;
        origin = cv::Point(x, y);
        selectRect = cv::Rect(x, y, 0, 0);
    }
    else if (event == CV_EVENT_LBUTTONUP)
    {
        select_flag = false;
        bRenewROI = true;
    }
}

当按下鼠标左键时，这个点就是起始点，按住鼠标左键移动鼠标，会选择感兴趣区域，松开鼠标左键，bRenewROI = true;修改标志，表示新的roi区域selectRect已经产生。在imageCb中程序继续执行，初始化KCFTracker，开始追踪。
到这里基本的流程已经比较清晰了，接下来开始将行人检测代替手动选择roi区域。

preparePeopleDetect()函数是初始化检测，
peopleDetect()函数是开始检测。

void preparePeopleDetect()
    {
        has_dectect_people = false;
        //hog.setSVMDetector(HOGDescriptor::getDefaultPeopleDetector());//使用默认的训练数据，下面是使用自己的训练数据。
        MySVM svm;
        string path = ros::package::getPath("track_pkg")+"/src/12000neg_2400pos.xml";
        printf("path === %s",path.c_str());
        //svm.load("/home/server/catkin_ws/src/tracker_kcf_ros/src/track_pkg/src/12000neg_2400pos.xml");
        svm.load(path.c_str());
        DescriptorDim = svm.get_var_count();//特征向量的维数，即HOG描述子的维数
        int supportVectorNum = svm.get_support_vector_count();//支持向量的个数
        cout<<"支持向量个数："<<supportVectorNum<<endl;

        Mat alphaMat = Mat::zeros(1, supportVectorNum, CV_32FC1);//alpha向量，长度等于支持向量个数
        Mat supportVectorMat = Mat::zeros(supportVectorNum, DescriptorDim, CV_32FC1);//支持向量矩阵
        Mat resultMat = Mat::zeros(1, DescriptorDim, CV_32FC1);//alpha向量乘以支持向量矩阵的结果

        //将支持向量的数据复制到supportVectorMat矩阵中
        for(int i=0; i<supportVectorNum; i++)
        {
            const float * pSVData = svm.get_support_vector(i);//返回第i个支持向量的数据指针
            for(int j=0; j<DescriptorDim; j++)
            {
                supportVectorMat.at<float>(i,j) = pSVData[j];
            }
        }

        //将alpha向量的数据复制到alphaMat中
        double * pAlphaData = svm.get_alpha_vector();//返回SVM的决策函数中的alpha向量
        for(int i=0; i<supportVectorNum; i++)
        {
            alphaMat.at<float>(0,i) = pAlphaData[i];
        }

        //计算-(alphaMat * supportVectorMat),结果放到resultMat中
        //gemm(alphaMat, supportVectorMat, -1, 0, 1, resultMat);//不知道为什么加负号？
        resultMat = -1 * alphaMat * supportVectorMat;

        //得到最终的setSVMDetector(const vector<float>& detector)参数中可用的检测子

        //将resultMat中的数据复制到数组myDetector中
        for(int i=0; i<DescriptorDim; i++)
        {
            myDetector.push_back(resultMat.at<float>(0,i));
        }
        //最后添加偏移量rho，得到检测子
        myDetector.push_back(svm.get_rho());
        cout<<"检测子维数："<<myDetector.size()<<endl;
        hog.setSVMDetector(myDetector);
        ofstream fout("HOGDetectorForOpenCV.txt");
        for(int i=0; i<myDetector.size(); i++)
        {
            fout<<myDetector[i]<<endl;
        }
        printf("Start the tracking process\\n");

    }
    //行人检测
    void peopleDetect()
    {
        if(has_dectect_people)
            return;
        vector<Rect> found, found_filtered;
        double t = (double)getTickCount();
        hog.detectMultiScale(rgbimage, found, 0, Size(8,8), Size(32,32), 1.05, 2);
        t = (double)getTickCount() - t;
        //printf("tdetection time = %gms\\n", t*1000./cv::getTickFrequency());
        size_t i, j;
        printf("found.size==%d",found.size());
        for( i = 0; i < found.size(); i++ )
        {
            Rect r = found[i];
            for( j = 0; j < found.size(); j++ )
                if( j != i && (r & found[j]) == r)
                    break;
            if( j == found.size() )
                found_filtered.push_back(r);
        }
        Rect r ;
        for( i = 0; i < found_filtered.size(); i++ )
        {
            r = found_filtered[i];
            // the HOG detector returns slightly larger rectangles than the real objects.
            // so we slightly shrink the rectangles to get a nicer output.
            r.x += cvRound(r.width*0.1);
            r.width = cvRound(r.width*0.8);
            r.y += cvRound(r.height*0.07);
            r.height = cvRound(r.height*0.8);
            //rectangle(rgbimage, r.tl(), r.br(), cv::Scalar(0,255,0), 3);
            //printf("r.x==%d,y==%d,width==%d,height==%d\\n",r.x,r.y,r.width,r.height);
        }
        //防止误检测
        if(r.width>100&&r.height>350){
            has_dectect_people=true;
            selectRect.x = r.x+(r.width-roi_width)/2;
            selectRect.y = r.y+(r.height-roi_height)/2;
            selectRect.width = roi_width;
            selectRect.height = roi_height;
            printf("selectRect.x==%d,y==%d,width==%d,height==%d\\n",selectRect.x,selectRect.y,selectRect.width,selectRect.height);
        }//imshow(RGB_WINDOW, rgbimage);
    }

检测到人后，人所在的区域是一个矩形，我这里在矩形区域内取其中间100*100的矩形为感兴趣区域。检测到人后将has_dectect_people置为true，使其不会再次检测。设置bRenewROI = true;select_flag = true;
select_flag：当追踪目标未消失时，为true，消失时为false，与bRenewROI一起作为是否重新检测行人追踪的标记。

完整代码如下

#include <iostream>
#include <fstream>
#include <sstream>
#include <algorithm>
#include <dirent.h>
#include <math.h>
#include <ros/ros.h>
#include <ros/package.h>
#include <image_transport/image_transport.h>
#include <cv_bridge/cv_bridge.h>
#include <sensor_msgs/image_encodings.h>
#include "geometry_msgs/Twist.h"

#include <opencv2/core/core.hpp>
#include <opencv2/highgui/highgui.hpp>
#include "opencv2/imgproc/imgproc.hpp"
#include "opencv2/objdetect/objdetect.hpp"
#include <stdio.h>
#include <string.h>
#include <ctype.h>
#include "kcftracker.hpp"

using namespace cv;
using namespace std;
static const std::string RGB_WINDOW = "RGB Image window";
//static const std::string DEPTH_WINDOW = "DEPTH Image window";

#define Max_linear_speed 1
#define Min_linear_speed 0.4
#define Min_distance 1.0
#define Max_distance 5.0
#define Max_rotation_speed 0.75

float linear_speed = 0;
float rotation_speed = 0;

float k_linear_speed = (Max_linear_speed - Min_linear_speed) / (Max_distance - Min_distance);
float h_linear_speed = Min_linear_speed - k_linear_speed * Min_distance;

float k_rotation_speed = 0.004;
float h_rotation_speed_left = 1.2;
float h_rotation_speed_right = 1.36;
float distance_scale = 1.0;
int ERROR_OFFSET_X_left1 = 100;
int ERROR_OFFSET_X_left2 = 300;
int ERROR_OFFSET_X_right1 = 340;
int ERROR_OFFSET_X_right2 = 600;
int roi_height = 100;
int roi_width = 100;
cv::Mat rgbimage;
cv::Mat depthimage;
cv::Rect selectRect;
cv::Point origin;
cv::Rect result;

bool select_flag = false;
bool bRenewROI = false;  // the flag to enable the implementation of KCF algorithm for the new chosen ROI
bool bBeginKCF = false;
bool enable_get_depth = false;

bool HOG = true;
bool FIXEDWINDOW = false;
bool MULTISCALE = true;
bool SILENT = true;
bool LAB = false;
int DescriptorDim;
bool has_dectect_people ;
// Create KCFTracker object
KCFTracker tracker(HOG, FIXEDWINDOW, MULTISCALE, LAB);
vector<float> myDetector;
float dist_val[5] ;

/*void onMouse(int event, int x, int y, int, void*)
{
    if (select_flag)
    {
        selectRect.x = MIN(origin.x, x);
        selectRect.y = MIN(origin.y, y);
        selectRect.width = abs(x - origin.x);
        selectRect.height = abs(y - origin.y);
        selectRect &= cv::Rect(0, 0, rgbimage.cols, rgbimage.rows);
    }
    if (event == CV_EVENT_LBUTTONDOWN)
    {
        bBeginKCF = false;
        select_flag = true;
        origin = cv::Point(x, y);
        selectRect = cv::Rect(x, y, 0, 0);
    }
    else if (event == CV_EVENT_LBUTTONUP)
    {
        select_flag = false;
        bRenewROI = true;
    }
}*/

class MySVM : public CvSVM
{
public:
    //获得SVM的决策函数中的alpha数组
    double * get_alpha_vector()
    {
        return this->decision_func->alpha;
    }

    //获得SVM的决策函数中的rho参数,即偏移量
    float get_rho()
    {
        return this->decision_func->rho;
    }
};

class ImageConverter
{
    ros::NodeHandle nh_;
    image_transport::ImageTransport it_;
    image_transport::Subscriber image_sub_;
    image_transport::Subscriber depth_sub_;
    HOGDescriptor hog;

public:
    ros::Publisher pub;

    ImageConverter()
    : it_(nh_)
    {
        // Subscrive to input video feed and publish output video feed
        image_sub_ = it_.subscribe("/camera/rgb/image_rect_color", 1,&ImageConverter::imageCb, this);
        depth_sub_ = it_.subscribe("/camera/depth/image", 1,&ImageConverter::depthCb, this);
        pub = nh_.advertise<geometry_msgs::Twist>("/mobile_base/mobile_base_controller/cmd_vel", 1000);
        //pub = nh_.advertise<geometry_msgs::Twist>("/cmd_vel", 1000);
        preparePeopleDetect();
        cv::namedWindow(RGB_WINDOW);
        //cv::namedWindow(DEPTH_WINDOW);
    }

    ~ImageConverter()
    {
        cv::destroyWindow(RGB_WINDOW);
        //cv::destroyWindow(DEPTH_WINDOW);
    }

    void imageCb(const sensor_msgs::ImageConstPtr& msg)
    {
        cv_bridge::CvImagePtr cv_ptr;
        try
        {
            cv_ptr = cv_bridge::toCvCopy(msg, sensor_msgs::image_encodings::BGR8);
        }
        catch (cv_bridge::Exception& e)
        {
            ROS_ERROR("cv_bridge exception: %s", e.what());
            return;
        }

        cv_ptr->image.copyTo(rgbimage);
        peopleDetect();
        if(has_dectect_people&&!select_flag)
        {
            printf("has_dectect_people = true \\n");
            selectRect &= cv::Rect(0,0,rgbimage.cols,rgbimage.rows);
            bRenewROI = true;
            select_flag = true;
        }
        //cv::setMouseCallback(RGB_WINDOW, onMouse, 0);
        if(bRenewROI)
        {
            // if (selectRect.width <= 0 || selectRect.height <= 0)
            // {
            //     bRenewROI = false;
            //     //continue;
            // }
            tracker.init(selectRect, rgbimage);
            bBeginKCF = true;
            bRenewROI = false;
            enable_get_depth = false;
        }
        if(bBeginKCF)
        {
            result = tracker.update(rgbimage);
            cv::rectangle(rgbimage, result, cv::Scalar( 0, 255,0 ), 1, 8 );
            enable_get_depth = true;
        }
        else
            cv::rectangle(rgbimage, selectRect, cv::Scalar(0, 255, 0), 2, 8, 0);

        cv::imshow(RGB_WINDOW, rgbimage);
        cv::waitKey(1);
    }
    void preparePeopleDetect()
    {
        has_dectect_people = false;
        //hog.setSVMDetector(HOGDescriptor::getDefaultPeopleDetector());
        MySVM svm;
        string path = ros::package::getPath("track_pkg")+"/src/12000neg_2400pos.xml";
        printf("path === %s",path.c_str());
        //svm.load("/home/server/catkin_ws/src/tracker_kcf_ros/src/track_pkg/src/12000neg_2400pos.xml");
        svm.load(path.c_str());
        DescriptorDim = svm.get_var_count();//特征向量的维数，即HOG描述子的维数
        int supportVectorNum = svm.get_support_vector_count();//支持向量的个数
        cout<<"支持向量个数："<<supportVectorNum<<endl;

        Mat alphaMat = Mat::zeros(1, supportVectorNum, CV_32FC1);//alpha向量，长度等于支持向量个数
        Mat supportVectorMat = Mat::zeros(supportVectorNum, DescriptorDim, CV_32FC1);//支持向量矩阵
        Mat resultMat = Mat::zeros(1, DescriptorDim, CV_32FC1);//alpha向量乘以支持向量矩阵的结果

        //将支持向量的数据复制到supportVectorMat矩阵中
        for(int i=0; i<supportVectorNum; i++)
        {
            const float * pSVData = svm.get_support_vector(i);//返回第i个支持向量的数据指针
            for(int j=0; j<DescriptorDim; j++)
            {
                supportVectorMat.at<float>(i,j) = pSVData[j];
            }
        }

        //将alpha向量的数据复制到alphaMat中
        double * pAlphaData = svm.get_alpha_vector();//返回SVM的决策函数中的alpha向量
        for(int i=0; i<supportVectorNum; i++)
        {
            alphaMat.at<float>(0,i) = pAlphaData[i];
        }

        //计算-(alphaMat * supportVectorMat),结果放到resultMat中
        //gemm(alphaMat, supportVectorMat, -1, 0, 1, resultMat);//不知道为什么加负号？
        resultMat = -1 * alphaMat * supportVectorMat;

        //得到最终的setSVMDetector(const vector<float>& detector)参数中可用的检测子

        //将resultMat中的数据复制到数组myDetector中
        for(int i=0; i<DescriptorDim; i++)
        {
            myDetector.push_back(resultMat.at<float>(0,i));
        }
        //最后添加偏移量rho，得到检测子
        myDetector.push_back(svm.get_rho());
        cout<<"检测子维数："<<myDetector.size()<<endl;
        hog.setSVMDetector(myDetector);
        ofstream fout("HOGDetectorForOpenCV.txt");
        for(int i=0; i<myDetector.size(); i++)
        {
            fout<<myDetector[i]<<endl;
        }
        printf("Start the tracking process\\n");

    }
    //行人检测
    void peopleDetect()
    {
        if(has_dectect_people)
            return;
        vector<Rect> found, found_filtered;
        double t = (double)getTickCount();
        hog.detectMultiScale(rgbimage, found, 0, Size(8,8), Size(32,32), 1.05, 2);
        t = (double)getTickCount() - t;
        //printf("tdetection time = %gms\\n", t*1000./cv::getTickFrequency());
        size_t i, j;
        printf("found.size==%d",found.size());
        for( i = 0; i < found.size(); i++ )
        {
            Rect r = found[i];
            for( j = 0; j < found.size(); j++ )
                if( j != i && (r & found[j]) == r)
                    break;
            if( j == found.size() )
                found_filtered.push_back(r);
        }
        Rect r ;
        for( i = 0; i < found_filtered.size(); i++ )
        {
            r = found_filtered[i];
            // the HOG detector returns slightly larger rectangles than the real objects.
            // so we slightly shrink the rectangles to get a nicer output.
            r.x += cvRound(r.width*0.1);
            r.width = cvRound(r.width*0.8);
            r.y += cvRound(r.height*0.07);
            r.height = cvRound(r.height*0.8);
            //rectangle(rgbimage, r.tl(), r.br(), cv::Scalar(0,255,0), 3);
            //printf("r.x==%d,y==%d,width==%d,height==%d\\n",r.x,r.y,r.width,r.height);
        }
        if(r.width>100&&r.height>350){
            has_dectect_people=true;
            selectRect.x = r.x+(r.width-roi_width)/2;
            selectRect.y = r.y+(r.height-roi_height)/2;
            selectRect.width = roi_width;
            selectRect.height = roi_height;
            printf("selectRect.x==%d,y==%d,width==%d,height==%d\\n",selectRect.x,selectRect.y,selectRect.width,selectRect.height);
        }//imshow(RGB_WINDOW, rgbimage);
    }
    void depthCb(const sensor_msgs::ImageConstPtr& msg)
    {
        cv_bridge::CvImagePtr cv_ptr;
        try
        {
            cv_ptr = cv_bridge::toCvCopy(msg,sensor_msgs::image_encodings::TYPE_32FC1);
            cv_ptr->image.copyTo(depthimage);
        }
        catch (cv_bridge::Exception& e)
        {
            ROS_ERROR("Could not convert from '%s' to 'TYPE_32FC1'.", msg->encoding.c_str());
        }

        if(enable_get_depth)
        {
            dist_val[0] = depthimage.at<float>(result.y+result.height/3 , result.x+result.width/3) ;
            dist_val[1] = depthimage.at<float>(result.y+result.height/3 , result.x+2*result.width/3) ;
            dist_val[2] = depthimage.at<float>(result.y+2*result.height/3 , result.x+result.width/3) ;
            dist_val[3] = depthimage.at<float>(result.y+2*result.height/3 , result.x+2*result.width/3) ;
            dist_val[4] = depthimage.at<float>(result.y+result.height/2 , result.x+result.width/2) ;

            float distance = 0;
            int num_depth_points = 5;
            for(int i = 0; i < 5; i++)
            {
                if(dist_val[i] > 0.4 && dist_val[i] < 10.0)
                    distance += dist_val[i];
                else
                    num_depth_points--;
            }
            distance /= num_depth_points;

            //calculate linear speed
            if(distance > Min_distance)
                linear_speed = distance * k_linear_speed + h_linear_speed;
            else if (distance <= Min_distance-0.5){
                //linear_speed = 0;
                linear_speed =-1* ((Min_distance-0.5) * k_linear_speed + h_linear_speed);
            }else{
                linear_speed = 0;
            }

            if( fabs(linear_speed) > Max_linear_speed)
                linear_speed = Max_linear_speed;

            //calculate rotation speed
            int center_x = result.x + result.width/2;
            if(center_x < ERROR_OFFSET_X_left1){
                printf("center_x <<<<<<<< ERROR_OFFSET_X_left1\\n");
                rotation_speed =  Max_rotation_speed/5;
                has_dectect_people = false;
                enable_get_depth = false;
                select_flag = false;
                bBeginKCF = false;
            }
            else if(center_x > ERROR_OFFSET_X_left1 && center_x < ERROR_OFFSET_X_left2)
                rotation_speed = -k_rotation_speed * center_x + h_rotation_speed_left;
            else if(center_x > ERROR_OFFSET_X_right1 && center_x < ERROR_OFFSET_X_right2)
                rotation_speed = -k_rotation_speed * center_x + h_rotation_speed_right;
            else if(center_x > ERROR_OFFSET_X_right2){
                printf("center_x >>>>>>>> ERROR_OFFSET_X_right2\\n");
                rotation_speed = -Max_rotation_speed/5;
                has_dectect_people = false;
                enable_get_depth = false;
                select_flag = false;
                bBeginKCF = false;
            }
            else
                rotation_speed = 0;

            //std::cout <<  "linear_speed = " << linear_speed << "  rotation_speed = " << rotation_speed << std::endl;

            // std::cout <<  dist_val[0]  << " / " <<  dist_val[1] << " / " << dist_val[2] << " / " << dist_val[3] <<  " / " << dist_val[4] << std::endl;
            // std::cout <<  "distance = " << distance << std::endl;
        }

        //cv::imshow(DEPTH_WINDOW, depthimage);
        cv::waitKey(1);
    }
};

int main(int argc, char** argv)
{
    ros::init(argc, argv, "kcf_tracker");
    ImageConverter ic;

    while(ros::ok())
    {
        ros::spinOnce();

        geometry_msgs::Twist twist;
        twist.linear.x = linear_speed;
        twist.linear.y = 0;
        twist.linear.z = 0;
        twist.angular.x = 0;
        twist.angular.y = 0;
        twist.angular.z = rotation_speed;
        ic.pub.publish(twist);
        if (cvWaitKey(33) == 'q')
            break;
    }

    return 0;
}

程序运行结果。