opencv knnSearch 使用

前言

本博客主要包括 利用 opencv 的最近邻搜索. 示例中包括 对一个点和对多个点的.
同时附带了自己目前经常用的Cmake 文件, 以及一个 python opencv 显示深度图的示例.

主要参考的网址有
OpenCV——KD Tree（介绍完整的flann邻近搜索）
opencv官网

主要的示例代码

#include
#include
#include
using namespace std;

int main(){
/*
 * vector m_pOriPtXY={ cv::Point2f(1,1), cv::Point2f(2,2),
               cv::Point2f(3,3), cv::Point2f(4,4), cv::Point2f(5,5), cv::Point2f(2,4)};
 * */
    vector<cv::Point2f> features={{1,1}, {2, 2}, {3, 3}, {4, 4}, {2, 4}};
    vector<cv::Point2f> raw_queries={{0,0}, {3, 4}, {4, 3}};

    /**建立kd树索引**/
    cv::Mat source = cv::Mat(features).reshape(1);
    //cout<
    source.convertTo(source,  CV_32F);
    //cout<
	// 注意这里其实最后还是转到一个 Mat 上 Mat 的大小是 n * p_dim [下面输出的size 是 先 width, 然后height]
    // 原本的意思是建立个新的地址进行转换
    cv::Mat queries = cv::Mat(raw_queries);
    //cout<
    queries = queries.reshape(1);
    //cout<
    queries.convertTo(queries,  CV_32F)
    //cout<

    cv::flann::KDTreeIndexParams indexParams(2);
    cv::flann::Index kdtree(source, indexParams); //此部分建立kd-tree索引同上例，故不做详细叙述

    /**预设knnSearch所需参数及容器**/
    unsigned queryNum = 2;//用于设置返回邻近点的个数
    vector<float> vecQuery(2);//存放 查询点 的容器（本例都是vector类型）
    vector<int> vecIndex(queryNum);//存放返回的点索引
    vector<float> vecDist(queryNum);//存放距离
    cv::flann::SearchParams params(32);//设置knnSearch搜索参数

    /**KD树knn查询**/
    vecQuery = {2.5, 2.6};
    kdtree.knnSearch(vecQuery, vecIndex, vecDist, queryNum, params);
    //请注意这句的逻辑：由先前生成的kdtree索引对象调用knnSearch()函数，进行点的knn搜索

    cv::Mat indices;
    cv::Mat dists;

    kdtree.knnSearch(queries, indices, dists, 1, params);

    cout<<"vecDist: "<<endl;
    for(auto&x :vecDist){
        cout<<x<<" ";
    }
    cout<<endl;

    cout<<"vecIndex: "<<endl;
    for(auto&x :vecIndex){
        cout<<x<<" ";
    }
    cout<<endl;

    cout<<indices<<endl; // 0 based
    cout<<dists<<endl; // pow
    return 0;
}

自己简单的CMake 模板

├── CMakeLists.txt
└── src/ # code in

cmake_minimum_required(VERSION 2.8) 
set(targetname test_ocv) 
PROJECT(${targetname}) 

SET(EXECUTABLE_OUTPUT_PATH ${PROJECT_SOURCE_DIR}/bin) 
SET(LIBRARY_OUTPUT_PATH ${PROJECT_SOURCE_DIR}/bin) 

# #####
if(CMAKE_COMPILER_IS_GNUCC)
    add_definitions(-std=c++11)
    set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} -pthread -O3")
endif(CMAKE_COMPILER_IS_GNUCC)
# #####

#######
if(DEFINED _chj_libs)
else()
    set(_chj_libs "")
endif()

IF (WIN32)
	include(xxxxxx/opencv3.1/OpenCVConfig.cmake)
ELSEIF (APPLE)
	MESSAGE(STATUS "Now is Apple systens.")
ELSEIF (UNIX)
    FIND_PACKAGE(OpenCV 3)
ENDIF ()

INCLUDE_DIRECTORIES(${OpenCV_INCLUDE_DIRS})
LINK_DIRECTORIES(${OPENCV_LIB_DIR})

LIST(APPEND _chj_libs opencv_core opencv_highgui opencv_imgproc opencv_calib3d opencv_imgcodecs )
######

file(GLOB_RECURSE CURRENT_HEADERS  ${CMAKE_CURRENT_SOURCE_DIR}/src/*.h)  
source_group("Include" FILES ${CURRENT_HEADERS})
file(GLOB_RECURSE Source  ${CMAKE_CURRENT_SOURCE_DIR}/src/*.cpp)  
source_group("Source" FILES ${Source})  

include_directories(
${CMAKE_CURRENT_SOURCE_DIR}/src
)

ADD_EXECUTABLE(main ${Source})

target_link_libraries(main ${_chj_libs} )

显示深度图

今天才知道不用plt, ocv也能做很多事情

with open(fdepth, "rb") as fp:
    raw = np.fromfile(fp, np.float32, h*w*3).reshape(h, w, 3)

raw = raw[:, :,2]
raw[raw>2000]=0 # 去掉背景

cv.normalize(raw, raw, 0,255, cv.NORM_MINMAX)
#p(raw.dtype)
raw=raw.astype(np.uint8)
#p(raw.dtype)
#new_raw=raw.copy()
raw = cv.applyColorMap(raw, cv.COLORMAP_JET)