认真的虎ORBSLAM2源码解读(十)：KeyFrameDataBase类

1.前言

2.头文件

class KeyFrameDatabase
{
public:

    KeyFrameDatabase(const ORBVocabulary &voc);

   void add(KeyFrame* pKF);

   void erase(KeyFrame* pKF);

   void clear();

   // Loop Detection
   //在KeyFrameDatabase，以及与pKF在covisibility graph连接的keyframe中找出与pKF可能形成闭环的候选帧
   //与DetectRelocalizationCandidates的区别是，在最开始先搜索了在covisibility graph与其连接的关键帧
   std::vector<KeyFrame *> DetectLoopCandidates(KeyFrame* pKF, float minScore);

   // Relocalization
   //在重定位时情形下，KeyFrameDatabase找出与当前帧相似F的候选关键帧，并返回
   std::vector<KeyFrame*> DetectRelocalizationCandidates(Frame* F);

protected:

  // Associated vocabulary
  const ORBVocabulary* mpVoc;

  // Inverted file
  //< 倒排索引，mvInvertedFile[i]表示包含了第i个word id的所有关键帧
  std::vector<list<KeyFrame*> > mvInvertedFile;

  // Mutex
  std::mutex mMutex;
};

3.源文件

3.1类方法DetectRelocalizationCandidates()

此方法返回与当前帧相似F的候选关键帧。

1.搜索所有和和F有着相同单词的keyframe存储在lKFsSharingWords；
2.从lKFsSharingWords中筛选出，相同单词数量较多的keyframe，计算通过dbow计算相似度得分存入lScoreAndMatch；
3.lScoreAndMatch中的keyframe，考虑其共视关系最高的keyframe然后再进行一次筛选得到AccScoreAndMatch；
4.第四次筛选得到lAccScoreAndMatch中所有得分超过0.75*bestAccScore的keyframe集合

vector<KeyFrame*> KeyFrameDatabase::DetectRelocalizationCandidates(Frame *F)
{
    //lKFsSharingWords存储着和F有着相同单词的keyframe
    list<KeyFrame*> lKFsSharingWords;

    // Search all keyframes that share a word with current frame
    //搜索所有和和F有着相同单词的keyframe存储在lKFsSharingWords
    //并且更新keyframe中mnRelocWords，表示和此F有多少共同的单词
    {
        unique_lock<mutex> lock(mMutex);

        for(DBoW2::BowVector::const_iterator vit=F->mBowVec.begin(), vend=F->mBowVec.end(); vit != vend; vit++)
        {
            list<KeyFrame*> &lKFs =   mvInvertedFile[vit->first];

            for(list<KeyFrame*>::iterator lit=lKFs.begin(), lend= lKFs.end(); lit!=lend; lit++)
            {
                KeyFrame* pKFi=*lit;
                if(pKFi->mnRelocQuery!=F->mnId)
                {
                    pKFi->mnRelocWords=0;
                    pKFi->mnRelocQuery=F->mnId;
                    lKFsSharingWords.push_back(pKFi);
                }
                pKFi->mnRelocWords++;
            }
        }
    }
    if(lKFsSharingWords.empty())
        return vector<KeyFrame*>();

    // Only compare against those keyframes that share enough words
    //在lKFsSharingWords中，寻找mnRelocWords的最大值存入maxCommonWords
    int maxCommonWords=0;
    for(list<KeyFrame*>::iterator lit=lKFsSharingWords.begin(), lend= lKFsSharingWords.end(); lit!=lend; lit++)
    {
        if((*lit)->mnRelocWords>maxCommonWords)
            maxCommonWords=(*lit)->mnRelocWords;
    }

    int minCommonWords = maxCommonWords*0.8f;

    list<pair<float,KeyFrame*> > lScoreAndMatch;

    int nscores=0;

    // Compute similarity score.
    //遍历lKFsSharingWords中的keyframe，当其中的keyframe的mRelocScore大于阈值minCommonWords则计算相似度后放入lScoreAndMatch中
    for(list<KeyFrame*>::iterator lit=lKFsSharingWords.begin(), lend= lKFsSharingWords.end(); lit!=lend; lit++)
    {
        KeyFrame* pKFi = *lit;

        if(pKFi->mnRelocWords>minCommonWords)
        {
            nscores++;
            float si = mpVoc->score(F->mBowVec,pKFi->mBowVec);
            pKFi->mRelocScore=si;
            lScoreAndMatch.push_back(make_pair(si,pKFi));
        }
    }

    if(lScoreAndMatch.empty())
        return vector<KeyFrame*>();

    list<pair<float,KeyFrame*> > lAccScoreAndMatch;
    float bestAccScore = 0;

    // Lets now accumulate score by covisibility
    //遍历lScoreAndMatch中的keyframe，找出其共视图中与此keyframe连接的权值前N的节点，加上原keyframe总共11个keyframe
    //累加这11个keyframe的相似度得分，然后在11个keyframe中选择相似度得分最高的那个放入lAccScoreAndMatch中
    //在遍历过程中计算bestAccScore，也就是AccScore的最大值，后面的再次筛选有用
    for(list<pair<float,KeyFrame*> >::iterator it=lScoreAndMatch.begin(), itend=lScoreAndMatch.end(); it!=itend; it++)
    {
        KeyFrame* pKFi = it->second;
	//返回共视图中与此keyframe连接的权值前10的节点keyframe
        vector<KeyFrame*> vpNeighs = pKFi->GetBestCovisibilityKeyFrames(10);

        float bestScore = it->first;
        float accScore = bestScore;
        KeyFrame* pBestKF = pKFi;
        for(vector<KeyFrame*>::iterator vit=vpNeighs.begin(), vend=vpNeighs.end(); vit!=vend; vit++)
        {
            KeyFrame* pKF2 = *vit;
	    //说明pKF2与F没有共同的单词，就放弃此循环的关键帧
            if(pKF2->mnRelocQuery!=F->mnId)
                continue;

            accScore+=pKF2->mRelocScore;
            if(pKF2->mRelocScore>bestScore)
            {
                pBestKF=pKF2;
                bestScore = pKF2->mRelocScore;
            }

        }
        lAccScoreAndMatch.push_back(make_pair(accScore,pBestKF));
        if(accScore>bestAccScore)
            bestAccScore=accScore;
    }

    // Return all those keyframes with a score higher than 0.75*bestAccScore
    //返回lAccScoreAndMatch中所有得分超过0.75*bestAccScore的keyframe集合
    float minScoreToRetain = 0.75f*bestAccScore;
    set<KeyFrame*> spAlreadyAddedKF;
    vector<KeyFrame*> vpRelocCandidates;
    vpRelocCandidates.reserve(lAccScoreAndMatch.size());
    for(list<pair<float,KeyFrame*> >::iterator it=lAccScoreAndMatch.begin(), itend=lAccScoreAndMatch.end(); it!=itend; it++)
    {
        const float &si = it->first;
        if(si>minScoreToRetain)
        {
            KeyFrame* pKFi = it->second;
            if(!spAlreadyAddedKF.count(pKFi))
            {
                vpRelocCandidates.push_back(pKFi);
                spAlreadyAddedKF.insert(pKFi);
            }
        }
    }

    return vpRelocCandidates;
}

3.2.DetectLoopCandidates()

   //在KeyFrameDatabase，以及与pKF在covisibility graph连接的keyframe中找出与pKF可能形成闭环的候选帧
   //与DetectRelocalizationCandidates的区别是，在最开始先搜索了在covisibility graph与其连接的关键帧

vector<KeyFrame*> KeyFrameDatabase::DetectLoopCandidates(KeyFrame* pKF, float minScore)
{
    //返回此关键帧在Covisibility graph中与之相连接（有共视关系）的节点
    set<KeyFrame*> spConnectedKeyFrames = pKF->GetConnectedKeyFrames();
    //和pKF有相同单词且具有共视关系的关键帧将会放在lKFsSharingWords
    list<KeyFrame*> lKFsSharingWords;

    // Search all keyframes that share a word with current keyframes
    // Discard keyframes connected to the query keyframe
    //找出与pKF有相同单词的关键帧
    {
        unique_lock<mutex> lock(mMutex);

	//遍历pKF中Bow里的每个单词
        for(DBoW2::BowVector::const_iterator vit=pKF->mBowVec.begin(), vend=pKF->mBowVec.end(); vit != vend; vit++)
        {
	    //通过单词id得到与之相有相同单词的关键帧
            list<KeyFrame*> &lKFs =   mvInvertedFile[vit->first];

	    //判断lKFs中关键帧是否复合要求
            for(list<KeyFrame*>::iterator lit=lKFs.begin(), lend= lKFs.end(); lit!=lend; lit++)
            {
                KeyFrame* pKFi=*lit;
		//如果pKFi被标记访问过
                if(pKFi->mnLoopQuery!=pKF->mnId)
                {
                    pKFi->mnLoopWords=0;
		    //如果pKFi和pKF有共视关系
                    if(!spConnectedKeyFrames.count(pKFi))
                    {
                        pKFi->mnLoopQuery=pKF->mnId;
                        lKFsSharingWords.push_back(pKFi);
                    }
                }
                //单词投票
                pKFi->mnLoopWords++;
            }
        }
    }

    if(lKFsSharingWords.empty())
        return vector<KeyFrame*>();

    list<pair<float,KeyFrame*> > lScoreAndMatch;

    // Only compare against those keyframes that share enough words
    //在lKFsSharingWords找出nLoopWords的最大值，也就是单词投票最多的哪个关键帧
    int maxCommonWords=0;
    for(list<KeyFrame*>::iterator lit=lKFsSharingWords.begin(), lend= lKFsSharingWords.end(); lit!=lend; lit++)
    {
        if((*lit)->mnLoopWords>maxCommonWords)
            maxCommonWords=(*lit)->mnLoopWords;
    }

    int minCommonWords = maxCommonWords*0.8f;

    int nscores=0;

    // Compute similarity score. Retain the matches whose score is higher than minScore
    //遍历lKFsSharingWords中的keyframe，当其中的keyframe的mLoopScore大于阈值minScore则计算相似度后放入lScoreAndMatch中
    for(list<KeyFrame*>::iterator lit=lKFsSharingWords.begin(), lend= lKFsSharingWords.end(); lit!=lend; lit++)
    {
        KeyFrame* pKFi = *lit;

        if(pKFi->mnLoopWords>minCommonWords)
        {
            nscores++;

            float si = mpVoc->score(pKF->mBowVec,pKFi->mBowVec);

            pKFi->mLoopScore = si;
            if(si>=minScore)
                lScoreAndMatch.push_back(make_pair(si,pKFi));
        }
    }

    if(lScoreAndMatch.empty())
        return vector<KeyFrame*>();

    list<pair<float,KeyFrame*> > lAccScoreAndMatch;
    float bestAccScore = minScore;

    // Lets now accumulate score by covisibility
    //遍历lScoreAndMatch中的keyframe，找出其共视图中与此keyframe连接的权值前N的节点，加上原keyframe总共最多有11个keyframe
    //累加这11个keyframe的相似度得分，然后在11个keyframe中选择相似度得分最高的那个放入lAccScoreAndMatch中
    //在遍历过程中计算bestAccScore，也就是AccScore的最大值，后面的再次筛选有用
    for(list<pair<float,KeyFrame*> >::iterator it=lScoreAndMatch.begin(), itend=lScoreAndMatch.end(); it!=itend; it++)
    {
        KeyFrame* pKFi = it->second;
	//返回共视图中与此keyframe连接的权值前10的节点keyframe
        vector<KeyFrame*> vpNeighs = pKFi->GetBestCovisibilityKeyFrames(10);

        float bestScore = it->first;
        float accScore = it->first;
        KeyFrame* pBestKF = pKFi;
        for(vector<KeyFrame*>::iterator vit=vpNeighs.begin(), vend=vpNeighs.end(); vit!=vend; vit++)
        {
            KeyFrame* pKF2 = *vit;
	    //如果pKF2与pKF有相同的单词，且在essentialgraph中连接，
	    //且pKF2的mLoopScore大于阈值minScore则将其相似度的值累加至accScore
            if(pKF2->mnLoopQuery==pKF->mnId && pKF2->mnLoopWords>minCommonWords)
            {
                accScore+=pKF2->mLoopScore;
                if(pKF2->mLoopScore>bestScore)
                {
                    pBestKF=pKF2;
                    bestScore = pKF2->mLoopScore;
                }
            }
        }

        lAccScoreAndMatch.push_back(make_pair(accScore,pBestKF));
        if(accScore>bestAccScore)
            bestAccScore=accScore;
    }

    // Return all those keyframes with a score higher than 0.75*bestScore
    float minScoreToRetain = 0.75f*bestAccScore;

    set<KeyFrame*> spAlreadyAddedKF;
    vector<KeyFrame*> vpLoopCandidates;
    vpLoopCandidates.reserve(lAccScoreAndMatch.size());

    //返回lAccScoreAndMatch中所有得分超过0.75*bestAccScore的keyframe集合
    for(list<pair<float,KeyFrame*> >::iterator it=lAccScoreAndMatch.begin(), itend=lAccScoreAndMatch.end(); it!=itend; it++)
    {
        if(it->first>minScoreToRetain)
        {
            KeyFrame* pKFi = it->second;
            if(!spAlreadyAddedKF.count(pKFi))
            {
                vpLoopCandidates.push_back(pKFi);
                spAlreadyAddedKF.insert(pKFi);
            }
        }
    }


    return vpLoopCandidates;
}