Learn opencv--particle filter (two articles on-line summary)

The theory of particle filtering is so wonderful that it approximates complex probability density functions with a set of random states of different weights. It has excellent properties in the nonlinear and non-Gaussian systems. OpenCV gives an implementation, but does not give an example, the learning process found on the network can not find. Learning OpenCV is described in a book, but it is still a distance away from direct use. After some bumpy, finally can use, hope to help you.

The example given in this article is one of the same applications as my other blog post, which is to smooth and predict two-dimensional coordinates.

How to use:

1. Create and initialize

const INT statenum=4;//State number
const int measurenum=2;//measurement variable number
const int samplenum=2000;//number of particles

cvcondensation* Condens = cvcreatecondensation (statenum,measurenum,samplenum);

The larger the number of particles, the more stable the system performance without compromising performance

Please refer to learning OPENCV for other initialization contents.

2. Forecasting
3. Update the example confidence, which is the weight. In this example, the Update method differs from learning OpenCV, and you want to see the code
4. Update Cvcondensation

Code:

[CPP]View PlainCopyprint?

2. #include <cv.h>
4. #include <cxcore.h>
6. #include
8. #include <cvaux.h>
12. #include <cmath>
14. #include <vector>
16. #include <iostream>
18. Using namespace std;
22. const int winheight=600;
24. const int winwidth=800;
30. Cvpoint Mouseposition=cvpoint (winwidth>>1,winheight>>1);
34. Mouse Event Callback
36. void mouseEvent (int event,int x,int y,int flags,void *param)
38. {
40. if (event==cv_event_mousemove) {
42. Mouseposition=cvpoint (x, y);
44. }
46. }
50. int main (void)
52. {
54. //1.condensation Setup
56. const int statenum=4;
58. const int measurenum=2;
60. const int samplenum=2000;
64. cvcondensation* Condens = cvcreatecondensation (statenum,measurenum,samplenum);
66. Cvmat* lowerbound;
68. cvmat* Upperbound;
70. Lowerbound = Cvcreatemat (Statenum, 1, cv_32f);
72. Upperbound = Cvcreatemat (Statenum, 1, cv_32f);
74. Cvmset (lowerbound,0,0,0.0);
76. Cvmset (Upperbound,0,0,winwidth);
78. Cvmset (lowerbound,1,0,0.0);
80. Cvmset (Upperbound,1,0,winheight);
82. Cvmset (lowerbound,2,0,0.0);
84. Cvmset (upperbound,2,0,0.0);
86. Cvmset (lowerbound,3,0,0.0);
88. Cvmset (upperbound,3,0,0.0);
90. float A[statenum][statenum] ={
92. 1,0,1,0,
94. 0,1,0,1,
96. 0,0,1,0,
98. 0,0,0,1
100. };
102. memcpy (Condens->dynammatr,a,sizeof (A));
104. Cvcondensinitsampleset (Condens, lowerbound, upperbound);
108. Cvrng rng_state = cvrng (0xFFFFFFFF);
110. For (int i=0; i < samplenum; i++) {
112. Condens->flsamples[i][0] = float (cvrandint (&rng_state)% winwidth); //width
114. CONDENS->FLSAMPLES[I][1] = float (cvrandint (&rng_state)% winheight); Height
116. }
120. Cvfont font;
122. Cvinitfont (&font,cv_font_hershey_script_complex,1,1);
126. char* winname="condensation";
128. Cvnamedwindow (Winname);
130. Cvsetmousecallback (winname,mouseevent);
132. iplimage* img=cvcreateimage (Cvsize (winwidth,winheight), 8, 3);
134. bool ispredictonly=false; Trigger for prediction only,press SPACEBAR
136. While (1) {
138. //2.condensation Prediction
140. Cvpoint predict_pt=cvpoint ((int) condens->state[0], (int) condens->state[1]);
144. float variance[measurenum]={0};
146. //get variance/standard Deviation of each State
148. For (int i=0;i<measurenum;i++) {
150. //sum
152. float sumstate=0;
154. For (int j=0;j<condens->samplesnum;j++) {
156. sumstate+=condens->flsamples[i][j];
158. }
160. //average
162. Sumstate/=samplenum;
164. //variance
166. For (int j=0;j<condens->samplesnum;j++) {
168. variance[i]+= (condens->flsamples[i][j]-sumstate) *
170. (condens->flsamples[i][j]-sumstate);
172. }
174. Variance[i]/=samplenum-1;
176. }
178. //3.update particals Confidence
180. Cvpoint pt;
182. if (ispredictonly) {
184. Pt=predict_pt;
186. }else{
188. Pt=mouseposition;
190. }
192. For (int i=0;i<condens->samplesnum;i++) {
194. float probx= (float) exp ( -1* (pt.x-condens->flsamples[i][0])
196. * (Pt.x-condens->flsamples[i][0])/(2*variance[0]));
198. float proby= (float) exp ( -1* (pt.y-condens->flsamples[i][1])
200. * (Pt.y-condens->flsamples[i][1])/(2*variance[1]));
202. condens->flconfidence[i]=probx*proby;
204. }
206. //4.update Condensation
208. Cvcondensupdatebytime (Condens);
212. //draw
214. Cvset (Img,cvscalar (255,255,255,0));
216. Cvcircle (Img,predict_pt,5,cv_rgb (0,255,0), 3); //predicted Point with green
218. Char buf[256];
220. sprintf_s (buf,256,"predicted position: (%3d,%3d)", PREDICT_PT.X,PREDICT_PT.Y);
222. Cvputtext (Img,buf,cvpoint (10,30), &font,cv_rgb (0,0,0));
224. if (!ispredictonly) {
226. Cvcircle (Img,mouseposition,5,cv_rgb (255,0,0), 3); //current position with red
228. sprintf_s (buf,256,"real position:(%3d,%3d)", MOUSEPOSITION.X,MOUSEPOSITION.Y);
230. Cvputtext (Img,buf,cvpoint (10,60), &font,cv_rgb (0,0,0));
232. }
236. Cvshowimage (Winname, IMG);
238. int Key=cvwaitkey (30);
240. if (key==27) {//esc
242. Break ;
244. }Else if (key==') {//trigger for prediction
246. //ispredict=!ispredict;
248. if (ispredictonly) {
250. Ispredictonly=false;
252. }else{
254. Ispredictonly=true;
256. }
258. }
260. }
264. Cvreleaseimage (&IMG);
266. Cvreleasecondensation (&condens);
268. return 0;
270. }

Kalman Filter Video Demo:

The number of particles in the demo is 100,200,2000

Please observe the effect carefully

Http://v.youku.com/v_show/id_XMjU4MzE0ODgw.html

Demo Snapshot:

the above article is a demonstration point tracking, the original http://blog.csdn.net/onezeros/article/details/6319180

This one is handed in to a Buddy demo window Tracker! (with code) http://www.cnblogs.com/yangyangcv/archive/2010/05/23/1742263.html from:http://blog.csdn.net/yangtrees/ article/details/7616483

Learn opencv--particle filter (two articles on-line summary)