sirflt_8hpp-source.html

Bayes滤波器算法C++类说明文档,源码见Bayes滤波器算法

<a name="l00142"></a><a class="code" href="classBayesian__filter_1_1Sampled__general__predict__model.html#a2">00142</a>     <span class="keywordtype">void</span> <a class="code" href="classBayesian__filter_1_1Sampled__general__predict__model.html#a2">init_GqG</a>() <span class="keyword">const</span>
00143     <span class="comment">/* initialise predict given a change to q,G</span>
00144 <span class="comment">     *  Implementation: Update rootq</span>
00145 <span class="comment">     */</span>
00146     {
00147         first_init = <span class="keyword">false</span>;
00148         <span class="keywordflow">for</span> (FM::Vec::const_iterator qi = this-&gt;q.begin(); qi != this-&gt;q.end(); ++qi) {
00149             <span class="keywordflow">if</span> (*qi &lt; 0.)
00150                 <a class="code" href="classBayesian__filter_1_1Bayes__base.html#e0">error</a> (<a class="code" href="classBayesian__filter_1_1Numeric__exception.html">Numeric_exception</a>(<span class="stringliteral">"Negative q in init_GqG"</span>));
00151             rootq[qi.index()] = std::sqrt(*qi);
00152         }
00153     }
00154 <span class="keyword">private</span>:
00155     SIR_random&amp; genn;
00156     <span class="keyword">mutable</span> FM::Vec xp;
00157     <span class="keyword">mutable</span> FM::DenseVec n;
00158     <span class="keyword">mutable</span> FM::Vec rootq;      <span class="comment">// Optimisation of sqrt(q) calculation, automatic on first use</span>
00159     <span class="keyword">mutable</span> <span class="keywordtype">bool</span> first_init;    
00160 };
00161 
<a name="l00162"></a><a class="code" href="namespaceBayesian__filter.html#a1">00162</a> <span class="keyword">typedef</span> <a class="code" href="classBayesian__filter_1_1Sampled__general__predict__model.html">Sampled_general_predict_model&lt;Linear_predict_model&gt;</a> <a class="code" href="namespaceBayesian__filter.html#a1">Sampled_LiAd_predict_model</a>;
<a name="l00163"></a><a class="code" href="namespaceBayesian__filter.html#a2">00163</a> <span class="keyword">typedef</span> <a class="code" href="classBayesian__filter_1_1Sampled__general__predict__model.html">Sampled_general_predict_model&lt;Linear_invertable_predict_model&gt;</a> <a class="code" href="namespaceBayesian__filter.html#a2">Sampled_LiInAd_predict_model</a>;
00164 <span class="comment">// Sampled predict model generalisations</span>
00165 <span class="comment">//  Names a shortened to first two letters of their model properties</span>
00166 
00167 
00168 
<a name="l00169"></a><a class="code" href="classBayesian__filter_1_1SIR__scheme.html">00169</a> <span class="keyword">class </span><a class="code" href="classBayesian__filter_1_1SIR__scheme.html">SIR_scheme</a> : <span class="keyword">public</span> <a class="code" href="classBayesian__filter_1_1Sample__filter.html">Sample_filter</a>
00170 <span class="comment">/*</span>
00171 <span class="comment"> * Sampling Importance Resampleing Filter Scheme.</span>
00172 <span class="comment"> *  Implement a general form of SIR filter</span>
00173 <span class="comment"> *  Importance resampling is delayed until an update is required. The sampler used</span>
00174 <span class="comment"> *  is a parameter of update to allow a wide variety of usage.</span>
00175 <span class="comment"> *  A stochastic sample is defined as a sample with a unqiue stochastic history other then roughening</span>
00176 <span class="comment"> */</span>
00177 {
<a name="l00178"></a><a class="code" href="classBayesian__filter_1_1SIR__scheme.html#n0">00178</a>     <span class="keyword">friend</span> <span class="keyword">class </span><a class="code" href="classBayesian__filter_1_1SIR__kalman__scheme.html">SIR_kalman_scheme</a>;
00179 <span class="keyword">public</span>:
<a name="l00180"></a><a class="code" href="classBayesian__filter_1_1SIR__scheme.html#o0">00180</a>     size_t <a class="code" href="classBayesian__filter_1_1SIR__scheme.html#o0">stochastic_samples</a>;  <span class="comment">// Number of stochastic samples in S</span>
00181 
00182     <a class="code" href="classBayesian__filter_1_1SIR__scheme.html#a0">SIR_scheme</a> (size_t x_size, size_t s_size, <a class="code" href="structBayesian__filter_1_1SIR__random.html">SIR_random</a>&amp; random_helper);
00183     <a class="code" href="classBayesian__filter_1_1SIR__scheme.html">SIR_scheme</a>&amp; <a class="code" href="classBayesian__filter_1_1SIR__scheme.html#a1">operator= </a>(<span class="keyword">const</span> <a class="code" href="classBayesian__filter_1_1SIR__scheme.html">SIR_scheme</a>&amp;);
00184     <span class="comment">// Optimise copy assignment to only copy filter state</span>
00185 
00186     <span class="comment">/* Specialisations for filter algorithm */</span>
00187     <span class="keywordtype">void</span> <a class="code" href="classBayesian__filter_1_1SIR__scheme.html#a2">init_S</a> ();
00188 
<a name="l00189"></a><a class="code" href="classBayesian__filter_1_1SIR__scheme.html#a3">00189</a>     <a class="code" href="classBayesian__filter_1_1Bayes__base.html#w0">Float</a> <a class="code" href="classBayesian__filter_1_1SIR__scheme.html#a3">update_resample</a> ()
00190     <span class="comment">// Default resampling update</span>
00191     {   <span class="keywordflow">return</span> <a class="code" href="classBayesian__filter_1_1SIR__scheme.html#a3">update_resample</a> (<a class="code" href="classBayesian__filter_1_1Standard__resampler.html">Standard_resampler</a>());
00192     }
00193 
00194     <span class="keyword">virtual</span> <a class="code" href="classBayesian__filter_1_1Bayes__base.html#w0">Float</a> <a class="code" href="classBayesian__filter_1_1SIR__scheme.html#a3">update_resample</a> (<span class="keyword">const</span> Importance_resampler&amp; resampler);
00195     <span class="comment">/* Update: resample particles using weights and then roughen</span>
00196 <span class="comment">     *  Return: lcond</span>
00197 <span class="comment">     */</span>
00198 
<a name="l00199"></a><a class="code" href="classBayesian__filter_1_1SIR__scheme.html#a5">00199</a>     <span class="keywordtype">void</span> <a class="code" href="classBayesian__filter_1_1SIR__scheme.html#a5">predict</a> (<a class="code" href="classBayesian__filter_1_1Functional__predict__model.html">Functional_predict_model</a>&amp; f)
00200     <span class="comment">// Predict samples without noise</span>
00201     {   Sample_filter::predict (f);
00202     }
00203     <span class="keywordtype">void</span> <a class="code" href="classBayesian__filter_1_1SIR__scheme.html#a5">predict</a> (Sampled_predict_model&amp; f);
00204     <span class="comment">// Predict samples with noise model</span>
00205 
00206     <span class="keywordtype">void</span> <a class="code" href="classBayesian__filter_1_1SIR__scheme.html#a7">observe</a> (Likelihood_observe_model&amp; h, <span class="keyword">const</span> FM::Vec&amp; z);
00207     <span class="comment">// Weight particles using likelihood model h and z</span>
00208 
00209     <span class="keywordtype">void</span> <a class="code" href="classBayesian__filter_1_1SIR__scheme.html#a8">observe_likelihood</a> (<span class="keyword">const</span> FM::Vec&amp; lw);
00210     <span class="comment">// Observation fusion directly from likelihood weights</span>
00211 
<a name="l00212"></a><a class="code" href="classBayesian__filter_1_1SIR__scheme.html#o1">00212</a>     <a class="code" href="classBayesian__filter_1_1Bayes__base.html#w0">Float</a> <a class="code" href="classBayesian__filter_1_1SIR__scheme.html#o1">rougheningK</a>;          <span class="comment">// Current roughening value (0 implies no roughening)</span>
<a name="l00213"></a><a class="code" href="classBayesian__filter_1_1SIR__scheme.html#a9">00213</a>     <span class="keyword">virtual</span> <span class="keywordtype">void</span> <a class="code" href="classBayesian__filter_1_1SIR__scheme.html#a9">roughen</a>()
00214     <span class="comment">// Generalised roughening:  Default to roughen_minmax</span>
00215     {
00216         <span class="keywordflow">if</span> (<a class="code" href="classBayesian__filter_1_1SIR__scheme.html#o1">rougheningK</a> != 0.)
00217             <a class="code" href="classBayesian__filter_1_1SIR__scheme.html#b0">roughen_minmax</a> (<a class="code" href="classBayesian__filter_1_1Sample__state__filter.html#o0">S</a>, <a class="code" href="classBayesian__filter_1_1SIR__scheme.html#o1">rougheningK</a>);
00218     }
00219 
00220     <span class="keyword">static</span> <span class="keywordtype">void</span> <a class="code" href="classBayesian__filter_1_1SIR__scheme.html#e0">copy_resamples</a> (FM::ColMatrix&amp; P, <span class="keyword">const</span> Importance_resampler::Resamples_t&amp; presamples);
00221     <span class="comment">// Update P by selectively copying based on presamples </span>
00222 
<a name="l00223"></a><a class="code" href="classBayesian__filter_1_1SIR__scheme.html#o2">00223</a>     <a class="code" href="structBayesian__filter_1_1SIR__random.html">SIR_random</a>&amp; <a class="code" href="classBayesian__filter_1_1SIR__scheme.html#o2">random</a>;         <span class="comment">// Reference random number generator helper</span>
00224 
00225 <span class="keyword">protected</span>:
00226     <span class="keywordtype">void</span> <a class="code" href="classBayesian__filter_1_1SIR__scheme.html#b0">roughen_minmax</a> (FM::ColMatrix&amp; P, Float K) <span class="keyword">const</span>;  <span class="comment">// roughening using minmax of P distribution</span>
<a name="l00227"></a><a class="code" href="classBayesian__filter_1_1SIR__scheme.html#p0">00227</a>     Importance_resampler::Resamples_t <a class="code" href="classBayesian__filter_1_1SIR__scheme.html#p0">resamples</a>;        <span class="comment">// resampling counts</span>
<a name="l00228"></a><a class="code" href="classBayesian__filter_1_1SIR__scheme.html#p1">00228</a>     FM::DenseVec <a class="code" href="classBayesian__filter_1_1SIR__scheme.html#p1">wir</a>;           <span class="comment">// resamping weights</span>
<a name="l00229"></a><a class="code" href="classBayesian__filter_1_1SIR__scheme.html#p2">00229</a>     <span class="keywordtype">bool</span> <a class="code" href="classBayesian__filter_1_1SIR__scheme.html#p2">wir_update</a>;            <span class="comment">// weights have been updated requring a resampling on update</span>
00230 <span class="keyword">private</span>:
00231     <span class="keyword">static</span> <span class="keyword">const</span> <a class="code" href="classBayesian__filter_1_1Bayes__base.html#w0">Float</a> rougheningKinit;
00232     size_t x_size;
00233 };
00234 
00235 
<a name="l00236"></a><a class="code" href="classBayesian__filter_1_1SIR__kalman__scheme.html">00236</a> <span class="keyword">class </span><a class="code" href="classBayesian__filter_1_1SIR__kalman__scheme.html">SIR_kalman_scheme</a> : <span class="keyword">public</span> <a class="code" href="classBayesian__filter_1_1SIR__scheme.html">SIR_scheme</a>, <span class="keyword">virtual</span> <span class="keyword">public</span> <a class="code" href="classBayesian__filter_1_1Kalman__state__filter.html">Kalman_state_filter</a>
00237 <span class="comment">/*</span>
00238 <span class="comment"> * SIR implementation of a Kalman filter</span>
00239 <span class="comment"> *  Updates Kalman statistics of SIR_filter</span>
00240 <span class="comment"> *  These statistics are use to provide a specialised correlated roughening procedure</span>
00241 <span class="comment"> */</span>
00242 {
00243 <span class="keyword">public</span>:
00244     <a class="code" href="classBayesian__filter_1_1SIR__kalman__scheme.html#a0">SIR_kalman_scheme</a> (size_t x_size, size_t s_size, <a class="code" href="structBayesian__filter_1_1SIR__random.html">SIR_random</a>&amp; random_helper);
00245 
00246     <span class="comment">/* Specialisations for filter algorithm */</span>
00247 
00248     <span class="keywordtype">void</span> <a class="code" href="classBayesian__filter_1_1SIR__kalman__scheme.html#a1">init</a> ();
00249 
<a name="l00250"></a><a class="code" href="classBayesian__filter_1_1SIR__kalman__scheme.html#a2">00250</a>     <span class="keywordtype">void</span> <a class="code" href="classBayesian__filter_1_1SIR__kalman__scheme.html#a2">update</a> ()
00251     <span class="comment">// Implement Kalman_filter::update identically to SIR_scheme</span>
00252     {   (<span class="keywordtype">void</span>)SIR_scheme::update_resample();
00253     }
00254 
<a name="l00255"></a><a class="code" href="classBayesian__filter_1_1SIR__kalman__scheme.html#a3">00255</a>     <a class="code" href="classBayesian__filter_1_1Bayes__base.html#w0">Float</a> <a class="code" href="classBayesian__filter_1_1SIR__kalman__scheme.html#a3">update_resample</a> ()
00256     <span class="comment">// Implement identically to SIR_scheme</span>
00257     {   <span class="keywordflow">return</span> SIR_scheme::update_resample();
00258     }
00259 
00260     <a class="code" href="classBayesian__filter_1_1Bayes__base.html#w0">Float</a> <a class="code" href="classBayesian__filter_1_1SIR__kalman__scheme.html#a3">update_resample</a> (<span class="keyword">const</span> Importance_resampler&amp; resampler);
00261     <span class="comment">// Modified SIR_filter update implementation: update mean and covariance of sampled distribution</span>
00262 
00263     <span class="keywordtype">void</span> <a class="code" href="classBayesian__filter_1_1SIR__kalman__scheme.html#a5">update_statistics</a> ();
00264     <span class="comment">// Update kalman statistics without resampling</span>
00265 
<a name="l00266"></a><a class="code" href="classBayesian__filter_1_1SIR__kalman__scheme.html#a6">00266</a>     <span class="keywordtype">void</span> <a class="code" href="classBayesian__filter_1_1SIR__kalman__scheme.html#a6">roughen</a>()
00267     {   <span class="comment">// Specialised correlated roughening</span>
00268         <span class="keywordflow">if</span> (<a class="code" href="classBayesian__filter_1_1SIR__scheme.html#o1">rougheningK</a> != 0.)
00269             <a class="code" href="classBayesian__filter_1_1SIR__kalman__scheme.html#b0">roughen_correlated</a> (<a class="code" href="classBayesian__filter_1_1Sample__state__filter.html#o0">S</a>, <a class="code" href="classBayesian__filter_1_1SIR__scheme.html#o1">rougheningK</a>);
00270     }
00271 
00272 <span class="keyword">protected</span>:
00273     <span class="keywordtype">void</span> <a class="code" href="classBayesian__filter_1_1SIR__kalman__scheme.html#b0">roughen_correlated</a> (FM::ColMatrix&amp; P, Float K);    <span class="comment">// Roughening using covariance of P distribution</span>
<a name="l00274"></a><a class="code" href="classBayesian__filter_1_1SIR__kalman__scheme.html#p0">00274</a>     <a class="code" href="namespaceBayesian__filter.html#a1">Sampled_LiAd_predict_model</a> <a class="code" href="classBayesian__filter_1_1SIR__kalman__scheme.html#p0">roughen_model</a>;       <span class="comment">// roughening predict</span>
00275 <span class="keyword">private</span>:
00276     <span class="keyword">static</span> <a class="code" href="classBayesian__filter_1_1Bayes__base.html#w0">Float</a> scaled_vector_square(<span class="keyword">const</span> FM::Vec&amp; v, <span class="keyword">const</span> FM::SymMatrix&amp; S);
00277     <span class="keywordtype">void</span> mean();
00278 };
00279 
00280 
00281 }<span class="comment">//namespace</span>
00282 <span class="preprocessor">#endif</span>
</pre></div><hr size="1"><address style="align: right;"><small>Generated on Mon Feb 16 11:20:40 2004 for Bayes++ Bayesian Filtering Classes by
<a href="http://www.doxygen.org/index.html">
<img src="doxygen.png" alt="doxygen" align="middle" border=0 > 
</a>1.3.2 </small></address>
</body>
</html>