Reduce processing time on Matlab motion detection script.

We currently have a matlab script that detects motion in some long videos of ours (hours) and outputs those detections as a list with the start, end and total time of the periods of motion detected. The only problem is that the script we have is very slow.

I tried reducing the number of frames that were read, because the motions we're looking for are quite significant (they can easily be seen with the human eye), however it seems I can only skip 2-3 frames before no motion is detected. Is there anyway I could make this script process faster? I don't see why only looking at every 25th frame in the video shouldn't be adequate for motion detection when the movements occur over minutes and can easily be seen with the human eye?

The code we're currently running is:

for k = 2 : 4 : nframes
                                waitbar(k/nframes,h);

                                  img_o=read(obj, k);  
                                  img_o=imresize(img_o,[new_r new_c]);
                                  img_oo=img_o;
                                  a=size(img_o);
                                  if(length(a)>2)                                     
                                      img_o=rgb2gray(img_o);                          
                                  end

                                  col_1=img_o(:);
                                  col_m=mean(col_1(:));

                                  if(col_m<5)
                                      motion=2;
                                      mov_mat=[];

                                      if(fig_flag==1)
                                          close(fig)
                                      end
                                      close(h1); 
                                      return
                                  end

                                  d=abs(double(img_l) - double(img_o));           
                                  d=d>th;                                        

                                  d=bwareaopen(d,10);                             
                                  d=imfill(d,'holes');

                                  se=strel('disk',3);
                                  d=imdilate(d,se);

                                  s=round((1-(sensitivity/100))*200);
                                  d=bwareaopen(d,s);

                                  [r,c]=find(d==1);
                                  rec_flag=0;
                                  if ~(isempty(r))
                                      [lab,num]=bwlabel(d);
                                      for i=1:num
                                          [r,c]=find(lab==i);
                                          pt_1(i,:)=[min(c) min(r)];
                                          len(i,1)=max(r)-min(r);
                                          br(i,1)=max(c)-min(c);
                                          rec_flag=1;
                                      end
                                      mov_frames=mov_frames + 1;
                                      non_mov_flag=0;

                                      if(mov_frames==1)
                                          last_mov_frame=k/4;
                                      end
                                      m_flag=1;
                                  else
                                      if(m_flag==1)
                                          non_mov_flag=non_mov_flag + 1;
                                          if(non_mov_flag>200 || k==nframes)
                                              m_flag=0;
                                              final_mov_frame=(k-(non_mov_flag))/4;
                                              mov_frames=0;

                                              mov_time=final_mov_frame - last_mov_frame;
                                              video_time_1=round(last_mov_frame/freq);
                                              video_time_2=round(final_mov_frame/freq);
                                              video_time_3=round(mov_time/freq);

                                              total_mov_time=total_mov_time + video_time_3 + 1;

                                               vid_time_ch_1=compute_time_1(video_time_1);
                                               vid_time_ch_2=compute_time_1(video_time_2);
                                               vid_time_ch_3=compute_time_1(video_time_3);
                                               %display(strcat('from:',vid_time_ch_1,'to:',vid_time_ch_2,'duration:',vid_time_ch_3));

                                               mov_mat{qq,1}=last_mov_frame;
                                               mov_mat{qq,2}=final_mov_frame;
                                               mov_mat{qq,3}=vid_time_ch_1;
                                               mov_mat{qq,4}=vid_time_ch_2;
                                               mov_mat{qq,5}=vid_time_ch_3;

                                               qq=qq+1;

                                          end
                                      end
                                  end

                              if(strcmp(display_status,'on'))
                                  dis_count=dis_count+1;
                                      if(rec_flag==1 && dis_count>50)
                                          dis_count=0;
                                          figure(2)
                                          fig = get(groot,'CurrentFigure');
                                          subplot(1,2,1)
                                          imshow(img_oo);
                                          hold on
                                          for i=1:size(pt_1,1)
                                              rectangle('Position',[pt_1(i,:) br(i,1) len(i,1)],'EdgeColor','g');
                                          end
                                          hold off

                                          title(strcat('frame:',num2str(k/4)));
                                          subplot(1,2,2)
                                          imshow(d);
                                          title('ROI binary')
                                          drawnow

                                          fig_flag=1;
                                      end

                              end
                              img_l=img_o;

                              if(rec_flag==1)
                                  motion=1;
                              end

                              clear pt_1 len br
                  end

                  if(total_mov_time/qq<=2)
                      motion=0;
                  end

                  total_mov_time=compute_time_1(total_mov_time-1);
                  mov_mat{1,6}=total_mov_time;
                  mov_mat{1,7}=nframes;
                  mov_mat{1,8}=freq;

                  if(fig_flag==1)
                      close(fig)
                  end
                  close(h1); 
                  toc
                  close(h);
end

Any help would be great!

Answer 1

[Image Analyst]

When you detect the motion at some frame, then just say it started at some frame earlier. Like if you're always 3 frame lagging from what people see, and you detection motion at frame 435, just say it happened at frame 432.