版块导航: 正在加载中...

登录注册

应《网络安全法》要求，自2017年10月1日起，未进行实名认证将不得使用互联网跟帖服务。为保障您的帐号能够正常使用，请尽快对帐号进行手机号验证，感谢您的理解与支持！

24小时热门版块排行榜

返回列表

winter678

新虫 (初入文坛)

应助: 0 (幼儿园)
金币: 8
帖子: 7
在线: 5.3小时
虫号: 3442073
注册: 2014-09-26
专业: 社会心理学

[交流] 有做过多假设目标跟踪(mht)的吗？这个“剪枝”程序对应的是什么方法?

% Prune cuts less possible hypoes and corresponding state estimations 剪枝
% from cellHypo and cellTarg as well as take record of good estimations.
%
% Syntax:
% [cellEstm, cellHypo, cellTarg] = ...
%    Prune(cellHypo, cellTarg, M, N, t, LP)
%
% In:
% cellEstm - a cell array containing the state estimation for each
%    target. The structure of each cell is {idx, startTime, matX}.
%    The startTime is the time step on which the target is confirmed.一个包含状态估计为每个单元阵列
%的目标。每个单元的结构{ idx、开始时间、matX }。目标确认作为的时间作为开始时间。
% cellHypo - a cell array of size 1*(1+M+M^2+...+M^N). The structure
%    of each cell in the cell array is {asso, prob}, where asso is a
%    column vector denoting the assignment of a set of measurements;
%    prob is a scalar denoting narual log of likelihood of this hypo. cellHypo -单元阵列的大小是1 *(1 + M + M ^ 2 +…+ M ^ N)结构
% cellTarg - a cell array of the same size of cellHypo. Each cell in
%    the cell array is called oneCase which gives a set of estimations
%    for all targets according to the correspoding hypo in cellHypo.
%    The structure of each cell in oneCase is {idx, lifePoint, X, P}
%    describing one target.
% M - number of hypoes generated for one seed hypo 一个干生成M个枝
% N - scan depth%扫描深度
% t - current time step. Note MHT introduces time delay, so at time
%    step t we can only confirm things happened at step t-N+1.
% LP - the confirmed hypo may contains some tentative targets that are
%    not included in previous estimations (i.e. input cellEstm). If
%    the lifePoints of such targets are not less than LP, then they
%    are confirmed to be new targets.
%
% Out:
% cellEstm - updated cellEstm
% cellHypo - cellHypo after pruning
% cellTarg - cellTarg after pruning

function [cellEstm, cellHypo, cellTarg] = Prune...
(cellEstm, cellHypo, cellTarg, M, N, t, LP)

% use the latest probabilities as the standard to judge%%？？？到54行不懂啊
head = 2;
for i = 1 : N-1
head = head + M^i;
end
rear = head + M^N - 1;
arrayProb = cellfun(@(v) v{2}, cellHypo(head : rear));%为cell数组的每个cell执行指定的函数 fun可以是特殊函数或者句柄
x = find(arrayProb == min(arrayProb)); % -log(prob), so find minimum
chooseBranch = ceil(x/M^(N-1));
idx = 1;
chooseIdx = [];
for i = 1 : N
chooseIdx = [chooseIdx, ...
      idx+(chooseBranch-1)*M^(i-1)+1 : idx+chooseBranch*M^(i-1)];
idx = idx + M^i;
end

% pick up the right branch
cellHypo = cellHypo(chooseIdx);
cellTarg = cellTarg(chooseIdx);

% update cellEstm
confirmedCase = cellTarg{1}; % the root of the hypo tree
for i = 1 : size(confirmedCase, 2)
aTarg = confirmedCase{i};
idx = aTarg{1};
lifePoint = aTarg{2};
X = aTarg{3};
if lifePoint == 0
      continue;
else
      flg = find(cellfun(@(v) v{1}, cellEstm) == idx);
      if isempty(flg) % maybe a new target appears
         if lifePoint >= LP
            newTarg = {idx, t-N+1, X};
            cellEstm = [cellEstm, newTarg];
         end
      else % an existed target
         thisTarg = cellEstm{flg};
         thisTarg{3} = [thisTarg{3}, X];
         cellEstm{flg} = thisTarg;
      end
end
end

%%怎么实现的剪枝啊？求交流

回复此楼

» 本帖附件资源列表

欢迎监督和反馈：小木虫仅提供交流平台，不对该内容负责。
本内容由用户自主发布，如果其内容涉及到知识产权问题，其责任在于用户本人，如对版权有异议，请联系邮箱：xiaomuchong@tal.com
附件 1 : MHT最小demo程序matlab.zip

2016-06-14 16:30:15, 72.3 K

» 猜你喜欢

1楼 2016-06-14 16:32:03

已阅回复此楼关注TA 给TA发消息送TA红花 TA的回帖

相关版块跳转我要订阅楼主 winter678 的主题更新