Breif 错误，Brisk 描述符 opencv答案

【问题标题】：Error with Breif , Brisk descriptor opencvBreif 错误，Brisk 描述符 opencv
【发布时间】：2015-02-17 10:14:30
【问题描述】：

在打开的简历中使用 BREIF 描述符时遇到错误。错误在于 BREIF 无法为所有关键点提供描述符。它给出的关键点数量少于预期的描述符。我得到的错误如下。错误仅使用 breif 描述符。使用 SURF 和 SURF 描述符没有错误

OpenCV 错误：checkMasks 文件中的断言失败 (masks[i].rows == queryDescriptorsCount && mask[i].cols == trainDescCollection[i].rows && mask[i].type() == CV_8UC1) /home/shashank/Desktop/opencv-2.4.5/modules/features2d/src/matchers.cpp，第 259 行在抛出 'cv::Exception' 的实例后调用终止什么（）：/home/shashank/Desktop/opencv-2.4.5/modules/features2d/src/matchers.cpp:259：错误：（-215）掩码[i].rows == queryDescriptorsCount && 掩码[i]。 cols == trainDescCollection[i].rows && mask[i].type() == CV_8UC1 in function checkMasks

进一步询问我们得到
上一张图片中关键点的编号 957 下一个图像中的关键点编号 910 breif 描述符的大小 32X880。应该是 32X957 breif 描述符的大小 32X847。应该是 32X910 面具 [910 x 957]

代码如下

     cv::Ptr<FeatureDetector> detector=new GridAdaptedFeatureDetector(new SurfAdjuster(),keypt,noofbucketinwidth,noofbucketinlength);//surf


 //cv::Ptr<FeatureDetector> detector=new GridAdaptedFeatureDetector(new FastAdjuster(10,true),keypt,noofbucketinwidth,noofbucketinlength); //FAST

     detector->detect(img,keypointimage);
    cout<<"noof keypoint "<<keypointimage.size()<<endl;


    //for descriptor
    Mat descriptor;
     Ptr<DescriptorExtractor> extractdetector=DescriptorExtractor::create("SURF");
     Ptr<DescriptorExtractor> extractdetector=DescriptorExtractor::create("BRIEF");   
        extractdetector->compute( img, detectedpoint, descriptor);    

    //for matching
    if(FAST_H_prev.empty())
            FAST_H_prev = Mat::eye(3,3,CV_32FC1);

        std::vector<unsigned char> FAST_match_mask;

        if(!currentimagekeypoint.empty())
        {
            std::vector<cv::KeyPoint> test_kpts;
            warpKeypoints(FAST_H_prev.inv(), nextimagekeypoint, test_kpts);

           cv::Mat FAST_mask = windowedMatchingMask( test_kpts, currentimagekeypoint, 25, 25);
                matcher->match(Discriptorofnextimage,Discriptorofcurrentimage, FAST_matches, FAST_mask);
                      matches2points(currentimagekeypoint, nextimagekeypoint, FAST_matches, FAST_train_pts, FAST_query_pts);
            if(FAST_matches.size() > 5)
            {
                cv::Mat H = findHomography(FAST_train_pts, FAST_query_pts, RANSAC, 4, FAST_match_mask);
                if(countNonZero(Mat(FAST_match_mask)) > 15)
                    FAST_H_prev = H;
                else
                    FAST_H_prev = Mat::eye(3,3,CV_32FC1);
                               drawMatchesRelative(currentimagekeypoint, nextimagekeypoint, FAST_matches, currentimage, FAST_match_mask);

            }
        }
    else
    {
        FAST_H_prev = Mat::eye(3,3,CV_32FC1);
    }

【问题讨论】：

标签： opencv computer-vision feature-detection feature-descriptor

【解决方案1】：

在 OpenCV 实现中，Breif 描述符会移除离边界太近的关键点。

//Remove keypoints very close to the border
KeyPointsFilter::runByImageBorder(keypoints, image.size(), PATCH_SIZE/2 + KERNEL_SIZE/2);

一种解决方案可能是过滤代码中的关键点，然后提取描述符。

【讨论】：