任务完成情况
参考这篇博客 OpenCV3特征提取和暴力匹配方法 "surf特征提取+匹配"
这篇博客是在opencv3版本上做特征提取的，3.0版本后的surf使用方法和2.0版本有了不同，所以需要做修改，这里参考了其使用的方法重新写了特征匹配的代码。
修改后的tracking_node代码：
/******************************************************************************
*
* The MIT License (MIT)
*
* Copyright (c) 2018 Bluewhale Robot
*
* Permission is hereby granted, free of charge, to any person obtaining a copy
* of this software and associated documentation files (the "Software"), to deal
* in the Software without restriction, including without limitation the rights
* to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
* copies of the Software, and to permit persons to whom the Software is
* furnished to do so, subject to the following conditions:
*
* The above copyright notice and this permission notice shall be included in all
* copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
* AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
* OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
* SOFTWARE.
*
* Author: Randoms
*******************************************************************************/
#include "tracking_node.h"
//beginging
#include <iostream>
#include <stdio.h>
#include <opencv2/core/core.hpp>
#include <opencv2/highgui/highgui.hpp>
#include <opencv2/imgproc.hpp>
#include <opencv2/features2d/features2d.hpp>
#include <opencv2/xfeatures2d/nonfree.hpp>
#include <opencv2/calib3d.hpp>
#include <opencv2/flann.hpp>
#include <opencv2/xfeatures2d.hpp>
//ending
using namespace cv;
using namespace std;
using namespace cv::xfeatures2d;
using namespace XiaoqiangTrack;
sensor_msgs::Image last_frame;
XiaoqiangTrack::Tracker *tracker = NULL;
Rect2d body_rect;
ros::Publisher image_pub;
ros::Publisher target_pub;
std::mutex update_track_mutex;
int track_ok_flag = 0;
cv::Rect2d previous_body_rect;
cv::Rect2d body_track_rect;
sensor_msgs::Image get_one_frame() { return last_frame; }
void update_frame(const sensor_msgs::ImageConstPtr &new_frame) //更新帧
{
last_frame = *new_frame;
cv_bridge::CvImagePtr cv_ptr =
cv_bridge::toCvCopy(new_frame, "bgr8"); //图像格式转换
cv::Mat cv_image = cv_ptr->image;
if (tracker == NULL)
return;
unique_lock<mutex> lock(update_track_mutex);
previous_body_rect = body_rect;//将检测到的当前rect保存为previous_body_rect
track_ok_flag = tracker->updateFrame(cv_image, body_rect);
cv::rectangle(cv_image, body_rect, cv::Scalar(0, 255, 0));
image_pub.publish(cv_ptr->toImageMsg());
xiaoqiang_track::TrackTarget target;
target.x = body_rect.x + body_rect.width / 2;
target.y = body_rect.y + body_rect.height / 2;
target.width = body_track_rect.width;
target.height = body_track_rect.height;
if (track_ok_flag == 0) {
// send stop
target.x = 0;
target.y = 0;
}
target_pub.publish(target);//target.x,target.y是跟踪的点的坐标，kalman...
}
int main(int argc, char **argv) {
ros::init(argc, argv, "xiaoqiang_track_node"); // ros节点初始化
//在一个节点中开辟多个线程，构造时可以指定线程数如(4)，AsyncSpinner有start()和stop()函数
ros::AsyncSpinner spinner(4);
spinner.start();
ros::NodeHandle private_nh("~");
ros::Publisher talk_pub = private_nh.advertise<std_msgs::String>("text", 10);
image_pub = private_nh.advertise<sensor_msgs::Image>("processed_image", 10);
target_pub = private_nh.advertise<xiaoqiang_track::TrackTarget>("target", 10);
int watch_dog;
private_nh.param("watch_dog", watch_dog, 2);
ros::Subscriber image_sub = private_nh.subscribe("image", 10, update_frame);
PoseTracker *client;
std::string pose_tracker_type;
ros::param::param<std::string>("~pose_tracker", pose_tracker_type, "");
if (pose_tracker_type == "baidu") //判断跟踪类型：baidu track or body track
{
client = (PoseTracker *)new BaiduTrack(private_nh);
} else if (pose_tracker_type == "body_pose") {
client = (PoseTracker *)new BodyTrack(private_nh);
} else {
ROS_FATAL_STREAM("unknown pose tracker type " << pose_tracker_type);
ROS_FATAL_STREAM("supported pose trakers are body_pose and baidu");
exit(1);
}
std::string tracker_main_type; //定义主跟踪类型
std::string tracker_aided_type; //辅跟踪
ros::param::param<std::string>("~tracker_main", tracker_main_type, "");
ros::param::param<std::string>("~tracker_aided", tracker_aided_type, "");
tracker = new XiaoqiangTrack::Tracker(tracker_main_type,tracker_aided_type); //设置跟踪器
// 告诉用户站在前面
std_msgs::String words;
words.data = "请站在我前面";
talk_pub.publish(words);
// 提醒用户调整好距离
sensor_msgs::Image frame = get_one_frame(); //得到一帧图像
body_rect.x = -1;
body_rect.y = -1;
while (!ros::isShuttingDown()) {
if (frame.data.size() != 0) {
cv::Rect2d rect = client->getBodyRect(frame); //通过frame得到人体图像区域
if (rect.x <= 1 || rect.y <= 1) {
words.data = "我没有看到人,请站到我前面";
talk_pub.publish(words);
} else if (rect.x + rect.width / 2 > 440 ||
rect.x + rect.width / 2 < 200) {
body_rect = rect;
words.data = "请站到镜头中间来";
talk_pub.publish(words);
} else {
body_rect = rect;
words.data = "我看到人了,开始追踪";
talk_pub.publish(words);
break;
}
}
sleep(1);
frame = get_one_frame();
}
/*
经过分析代码，初步的想法是在这个位置加上特征提取方法和Opencv的特征匹配，思路是：
-
特征提取是从一帧图像中提取特征，想要提取的特征可以是ORB，FAST，SIFT，SURF等，上面的
- frame = get_one_frame()是获取最新的一帧图像，return last_frame
-
那么对于这一帧图像抽取想要的特征信息，得到特征点，保存检测到的目标特征，之后用来与再次检测时的图像来做匹配
-
当然这里是做特征提取，匹配是在跟踪丢失后，再次启动检测识别时，识别到多个目标，进行匹配
*/
/*fuck!sorry,i don't wanna say that,but i just lost everything i did the whole day because of clicking the
close table without save it! so let me start at the begining...
*/
//begining extract feature
//int minHessian = 2000;
//int hessianThreshold = 2000;
//SurfFeatureDetector detector(minHessian);
//detector = SURF::create(hessianThreshold);
Ptr<SURF> surf;
surf = SURF::create(800);
vector<KeyPoint>keypoint1, keypoint2;
frame = get_one_frame();
body_rect = client->getBodyRect(frame);
//image1 = resize(frame, rect.x:(rect.x+rect.width), rect.y:(rect.y+rect.higth))
Mat image1;
/*
Mat img1,img_1,image1;
CvRect rectInImage1;
rectInImage1 = cvRect(body_rect.x, body_rect.y,body_rect.width, body_rect.height);
CvSize size1;
size1.width = rectInImage1.width;
size1.height = rectInImage1.height;
//img1 = CvCreatImage(size1, frame->depth, frame->nChannels);
img1 = CvCreatImage(size1, IPL_DEPTH_8U, 3);
CvSetImageROI(frame, rectInImage1);
cvCopy(img1, frame);//img1是从frame上提取的目标框区域
cvtcolor(img1, img_1, CV_RGB2GRAY);
*/
//cvtcolor(frame, frame, CV_RGB2GRAY)
//检测特征点
//detector.detect(img_1, keypoint1)
surf->detectAndCompute(frame, Mat(), keypoint1, image1); //输出mat存放所有的特征点描述向量
//ending
// 告诉用户可以开始走了
sensor_msgs::Image tracking_frame = get_one_frame();
cv_bridge::CvImagePtr cv_ptr = cv_bridge::toCvCopy(tracking_frame, "bgr8");
cv::Mat cv_image = cv_ptr->image;
tracker->initTracker(cv_image, body_rect); // init
int repeat_count = 0;
int watch_dog_count = 0;
while (ros::ok()) {
usleep(1000 * 100); // 100ms
// 如果位置不变，则认为可能丢失
if (previous_body_rect == body_rect) {
repeat_count += 100;
if (repeat_count == 5 * 1000) //rect检测到的数据不变，且超过一定时间，判Lost
{
ROS_WARN_STREAM("Target not move, may lost");
repeat_count = 0;
track_ok_flag = 0; // heihei,flag=0 -> reset
}
} //这里判断跟丢
else
{
repeat_count = 0;
}
if (body_rect.width < 300 && body_rect.height < 300 && track_ok_flag == 2 &&
body_rect.height > body_rect.width) //确认检测到的rect符合这些要求
{
watch_dog_count += 100;
if (watch_dog_count <= watch_dog * 1000) // watch_dog=2
{
continue;
}
} //这里判断是否正确的给出rect
watch_dog_count = 0;
tracking_frame = get_one_frame(); //再次获得newframe
body_track_rect = client->getBodyRect(tracking_frame); // track
//usr code begining
if(track_ok_flag == 0)
{
Mat image2;
/*
Mat img2, img_2, image2;
CvRect rectInImage2;
rectInImage2 = cvRect(body_track_rect.x, body_track_rect.y,
body_track_rect.width, body_track_rect.height);
CvSize size2;
size2.width = rectInImage2.width;
size2.height = rectInImage2.height;
//img2 = CvCreatImage(size2, tracking_frame->depth, tracking_frame->nChannels);
img2 = CvCreatImage(size2, IPL_DEPTH_8U, 3);
CvSetImageROI(tracking_frame, rectInImage2);
cvCopy(img2, tracking_frame);//img2是从tracking_frame上提取的目标框区域
cvtcolor(img2, img_2, CV_RGB2GRAY);
*/
//cvtcolor(tracking_frame, tracking_frame, CV_RGB2GRAY);
//检测特征点
//detector.detect(img_2, keypoint2);
surf->detectAndCompute(tracking_frame, Mat(), keypoint2, image2);
//计算特征点描述子
//SurfDescriptorExtractor extractor;
//Mat descriptor1, descriptor2;
//extractor.compute(img1, keypoint1, descriptor1);
//extractor.compute(img2, keypoint2, descriptor2);
//使用flann匹配
FlannBasedMatcher matcher;
vector<DMatch>matches;
matcher.match(image1, image2, matches);//匹配结束
sort(matches.begin(), matches.end());
vector<DMatch>good_matches;
int ptsPairs = std::min(50, (int)(matches.size() * 0.15));
for (int i = 0; i < ptsPairs; i++)
{
good_matches.push_back(matches[i]);
}
double averagematch =0;
for(int i = 0; i < ptsPairs; i++)
{
averagematch = averagematch + (int)(good_matches[i]) / ptsPairs;
}
if(averagematch > 10.0)
continue;
/*
double max_dist = 0;
double min_dist = 100;
for(int i = 0; i < descriptor1.rows; i++)
{
double dist = matches[i].distance;
if(dist < min_dist)
min_dist = dist;
if(dist > max_dist)
max_dist = dist;
}//得到匹配结果中的最小距离和最大距离
//处理匹配结果：判断当前匹配的对象是否为目标，仅根据最大最小匹配距离，能否进行判断？
if(!(min_dist < 1.0 || (max_dist - min_dist) < 2.0))// matching failed
{
//tracking_frame = tracking_frame;
//body_track_rect = body_track_rect;
continue;
}
*/
}
//usr code ending
if (body_track_rect.x <= 1 || body_track_rect.y <= 1) //识别到的rect.x / .y小于1后跟踪停止
{
tracker->stop();
}
else
{
{
unique_lock<mutex> lock(update_track_mutex);
body_rect = body_track_rect;
cv_bridge::CvImagePtr cv_ptr =
cv_bridge::toCvCopy(tracking_frame, "bgr8");
cv::Mat cv_image = cv_ptr->image;
if (track_ok_flag == 0) //跟踪标志为0时跟踪复位
{
tracker->reset(cv_image, body_rect, true); // watch out the reset praram
}
else
tracker->reset(cv_image, body_rect); /// reset
}
}
}
}代码中的关于目标框区域裁剪的部分做了注释，因为裁剪的语句中CvCreatImage、CvSetImageROI等都无法使用，说这些函数在当前工作空间中找不到。。
尝试将裁剪的部分注释掉后，还有一些报错，不能懂其中的原因，报的错误长这样：
/home/sz/catkin_hf/src/xiaoqiang_track/src/tracking_node.cpp: In function ‘int main(int, char**)’:
/home/sz/catkin_hf/src/xiaoqiang_track/src/tracking_node.cpp:194:57: error: no matching function for call to ‘cv::xfeatures2d::SURF::detectAndCompute(sensor_msgs::Image&, cv::Mat, std::vector<cv::KeyPoint>&, cv::Mat&)’
surf->detectAndCompute(frame, Mat(), keypoint1, image1); //输出mat存放所有的特征点描述向量
^
In file included from /opt/ros/kinetic/include/opencv-3.3.1-dev/opencv2/calib3d.hpp:48:0,
from /opt/ros/kinetic/include/opencv-3.3.1-dev/opencv2/opencv.hpp:56,
from /home/sz/catkin_hf/src/xiaoqiang_track/include/tracking_node.h:31,
from /home/sz/catkin_hf/src/xiaoqiang_track/src/tracking_node.cpp:28:
/opt/ros/kinetic/include/opencv-3.3.1-dev/opencv2/features2d.hpp:198:26: note: candidate: virtual void cv::Feature2D::detectAndCompute(cv::InputArray, cv::InputArray, std::vector<cv::KeyPoint>&, cv::OutputArray, bool)
CV_WRAP virtual void detectAndCompute( InputArray image, InputArray mask,
^
/opt/ros/kinetic/include/opencv-3.3.1-dev/opencv2/features2d.hpp:198:26: note: no known conversion for argument 1 from ‘sensor_msgs::Image {aka sensor_msgs::Image_<std::allocator<void> >}’ to ‘cv::InputArray {aka const cv::_InputArray&}’
/home/sz/catkin_hf/src/xiaoqiang_track/src/tracking_node.cpp:256:66: error: no matching function for call to ‘cv::xfeatures2d::SURF::detectAndCompute(sensor_msgs::Image&, cv::Mat, std::vector<cv::KeyPoint>&, cv::Mat&)’
surf->detectAndCompute(tracking_frame, Mat(), keypoint2, image2);
^
In file included from /opt/ros/kinetic/include/opencv-3.3.1-dev/opencv2/calib3d.hpp:48:0,
from /opt/ros/kinetic/include/opencv-3.3.1-dev/opencv2/opencv.hpp:56,
from /home/sz/catkin_hf/src/xiaoqiang_track/include/tracking_node.h:31,
from /home/sz/catkin_hf/src/xiaoqiang_track/src/tracking_node.cpp:28:
/opt/ros/kinetic/include/opencv-3.3.1-dev/opencv2/features2d.hpp:198:26: note: candidate: virtual void cv::Feature2D::detectAndCompute(cv::InputArray, cv::InputArray, std::vector<cv::KeyPoint>&, cv::OutputArray, bool)
CV_WRAP virtual void detectAndCompute( InputArray image, InputArray mask,
^
/opt/ros/kinetic/include/opencv-3.3.1-dev/opencv2/features2d.hpp:198:26: note: no known conversion for argument 1 from ‘sensor_msgs::Image {aka sensor_msgs::Image_<std::allocator<void> >}’ to ‘cv::InputArray {aka const cv::_InputArray&}’
/home/sz/catkin_hf/src/xiaoqiang_track/src/tracking_node.cpp:282:54: error: invalid cast from type ‘__gnu_cxx::__alloc_traits<std::allocator<cv::DMatch> >::value_type {aka cv::DMatch}’ to type ‘int’
averagematch = averagematch + (int)(good_matches[i]) / ptsPairs;
^
xiaoqiang_track/CMakeFiles/tracking_node.dir/build.make:62: recipe for target 'xiaoqiang_track/CMakeFiles/tracking_node.dir/src/tracking_node.cpp.o' failed
make[2]: *** [xiaoqiang_track/CMakeFiles/tracking_node.dir/src/tracking_node.cpp.o] Error 1
CMakeFiles/Makefile2:1400: recipe for target 'xiaoqiang_track/CMakeFiles/tracking_node.dir/all' failed
make[1]: *** [xiaoqiang_track/CMakeFiles/tracking_node.dir/all] Error 2
Makefile:138: recipe for target 'all' failed
make: *** [all] Error 2
[1] detectAndCompute这个函数这里提示no matching function
[2] averagematch = averagematch + (int)(good_matches[i]) / ptsPairs;这句是将匹配的结果取前50个距离最小匹配特征点再取均值，用到了强制类型转换，问题的出现原因应该是good_matches类型为容器类向量，不能给它类型转换为int，也不能直接与整数进行运算，这个问题现在要解决是要考虑怎么利用匹配的结果，来做判断是否匹配成功的问题，匹配的结果是获得特征点匹配结果的容器类向量数据，怎么用这些数据做判断是否是符合要求的匹配。