def depthmatch(x,y,leftimage,rightimage,roi=80,buf=50,baseline=2.7,focal_length=80): """depthmatch function x,y : (int) pixel position of target in left image leftimage, rightimage : (IplImage) stereo images roi: (int) region of interest around x,y to use in matching buf: (int) buffer outside of a straight horizontal search for a match """ #print "Match",x,y info = cv.cvGetSize(leftimage) width = info.width height = info.height centerx = width/2 centery = height/2 (y1,x1,y2,x2) = (y-roi,x-roi,y+roi,x+roi) if y1<0: y1 = 0 if x1<0: x1 = 0 if y2>height: y2 = height if x2>width: x2 = width # copy subregion roi x roi template_rect = cv.cvRect(x1,y1,(x2-x1),(y2-y1)) template = cv.cvGetSubRect(leftimage, template_rect) #(y3,x3,y4,x4) = (y-roi-buf,x-roi-buf,y+roi+buf,width) # +/- 20 pixels in vertical direction, -20 to the right edge (y3,x3,y4,x4) = (y-roi-buf,0,y+roi+buf,x+roi+buf) # +/- buf pixels in vertical direction, +buf to the left edge if x3<0: x3 = 0 if y3<0: y3 = 0 if x4>=width: x4 = width-1 if y4>height: y4 = height #cv.cvSetImageROI(rightimage, (y3,x3,y4,x4)) rightsub_rect = cv.cvRect(x3,y3,(x4-x3),(y4-y3)) rightsub = cv.cvGetSubRect(rightimage, rightsub_rect) # result matrix should be (W - w + 1) x (H - h + 1) where WxH are template dimensions, wxh are rightsub dimensions W = x4-x3 H = y4-y3 w = x2-x1 h = y2-y1 resy = (y4-y3)-(y2-y1)+1 resx = (x4-x3)-(x2-x1)+1 resultmat = cv.cvCreateImage((resx, resy), 32, 1) cv.cvZero(resultmat) # match template image in a subportion of rightimage cv.cvMatchTemplate(rightsub, template, resultmat, cv.CV_TM_SQDIFF) min_val, max_val, min_point, max_point = cv.cvMinMaxLoc(resultmat) cv.cvNormalize(resultmat, resultmat, 1, 0, cv.CV_MINMAX) depth = plane2point(x-centerx, y-centery, x3+min_point.x+roi-centerx, y3+min_point.y+roi-centery, baseline, focal_length) #print "Found match at", min_point.x+x3, min_point.y+y3 return (depth, (x,y), (x3+min_point.x+roi, y3+min_point.y+roi))
def read(self): raw_thresh = self.thresh.read() cvt_red = cv.cvCreateImage(cv.cvSize(raw_thresh.width,raw_thresh.height),raw_thresh.depth,1) cv.cvSplit(raw_thresh,cvt_red,None,None,None) cvpt_min = cv.cvPoint(0,0) cvpt_max = cv.cvPoint(0,0) t = cv.cvMinMaxLoc(cvt_red,cvpt_min,cvpt_max) if cvpt_max.x == 0 and cvpt_max.y == 0 : return [] return [(cvpt_max.x,cvpt_max.y)]
def read(self): src = self.camera.read() thresh = self.thresh2pg.read() red = self.red2pg.read() raw_thresh = self.thresh.read() cvt_red = cv.cvCreateImage(cv.cvSize(raw_thresh.width,raw_thresh.height),raw_thresh.depth,1) cv.cvSplit(raw_thresh,cvt_red,None,None,None) cvpt_min = cv.cvPoint(0,0) cvpt_max = cv.cvPoint(0,0) t = cv.cvMinMaxLoc(cvt_red,cvpt_min,cvpt_max) return src,thresh,red,(cvpt_max.x,cvpt_max.y)
def matchTemplate(self, template, image): ''' matchTemplate(self, template, image): \ returns - correlation value of best match (b/w 0 & 1) \ top-left coord of template for the best match (cvPoint) \ ''' matchResultHeight = image.height-template.height+1 matchResultWidth = image.width-template.width+1 #print 'matchResultHeight: %d matchResultWidth %d'%(matchResultHeight, matchResultWidth) matchResult = cv.cvCreateMat(matchResultHeight, matchResultWidth, cv.CV_32FC1) cv.cvMatchTemplate(image, template, matchResult, cv.CV_TM_CCORR_NORMED) min_loc = cv.cvPoint(0,0) max_loc = cv.cvPoint(0,0) min_val, max_val = cv.cvMinMaxLoc(matchResult, min_loc, max_loc) return {'image': matchResult , 'max_val':max_val, 'max_loc':max_loc}
def matchTemplate(self, template, image): ''' matchTemplate(self, template, image): \ returns - correlation value of best match (b/w 0 & 1) \ top-left coord of template for the best match (cvPoint) \ ''' matchResultHeight = image.height - template.height + 1 matchResultWidth = image.width - template.width + 1 #print 'matchResultHeight: %d matchResultWidth %d'%(matchResultHeight, matchResultWidth) matchResult = cv.cvCreateMat(matchResultHeight, matchResultWidth, cv.CV_32FC1) cv.cvMatchTemplate(image, template, matchResult, cv.CV_TM_CCORR_NORMED) min_loc = cv.cvPoint(0, 0) max_loc = cv.cvPoint(0, 0) min_val, max_val = cv.cvMinMaxLoc(matchResult, min_loc, max_loc) return {'image': matchResult, 'max_val': max_val, 'max_loc': max_loc}
def read(self): frame = self.input.read() if self.debug : raw_frame = cv.cvCreateImage(cv.cvSize(frame.width,frame.height),frame.depth,frame.nChannels) cv.cvCopy(frame,raw_frame,None) self.raw_frame_surface=pygame.image.frombuffer(frame.imageData,(frame.width,frame.height),'RGB') if self.enabled : cvt_red = cv.cvCreateImage(cv.cvSize(frame.width,frame.height),frame.depth,1) cv.cvSplit(frame,None,None,cvt_red,None) if self.debug : red_frame = cv.cvCreateImage(cv.cvSize(cvt_red.width,cvt_red.height),cvt_red.depth,3) cv.cvMerge(cvt_red,None,None,None,red_frame) self.red_frame_surface = pygame.image.frombuffer(red_frame.imageData,(cvt_red.width,cvt_red.height),'RGB') # I think these functions are too specialized for transforms cv.cvSmooth(cvt_red,cvt_red,cv.CV_GAUSSIAN,3, 0, 0, 0 ) cv.cvErode(cvt_red, cvt_red, None, 1) cv.cvDilate(cvt_red, cvt_red, None, 1) if self.debug : thresh_frame = cv.cvCreateImage(cv.cvSize(cvt_red.width,cvt_red.height),cvt_red.depth,3) cv.cvMerge(cvt_red,None,None,None,thresh_frame) self.thresh_frame_surface = pygame.image.frombuffer(cvt_red.imageData,(cvt_red.width,cvt_red.height),'RGB') cvpt_min = cv.cvPoint(0,0) cvpt_max = cv.cvPoint(0,0) t = cv.cvMinMaxLoc(cvt_red,cvpt_min,cvpt_max) print t if cvpt_max.x == 0 and cvpt_max.y == 0 : return [] return [(cvpt_max.x,cvpt_max.y)]
def depthmatch(x,y,leftimage,rightimage,roi=20,buf=10,debug=False): __doc__ = """depthmatch function x,y : (int) pixel position of target in left image leftimage, rightimage : (IplImage) stereo images roi: (int) region of interest around x,y to use in matching buf: (int) buffer outside of a straight horizontal search for a match """ info = cv.cvGetSize(leftimage) width = info.width height = info.height (y1,x1,y2,x2) = (y-roi,x-roi,y+roi,x+roi) #template = cv.cvCreateImage((roi*2,roi*2), 8, 3) if y1<0: y1 = 0 if x1<0: x1 = 0 if y2>height: y2 = height if x2>width: x2 = width #cv.cvSetZero(template) # copy subregion roi x roi template_rect = cv.cvRect(x1,y1,(x2-x1),(y2-y1)) template = cv.cvGetSubRect(leftimage, template_rect) (y3,x3,y4,x4) = (y-roi-buf,x-roi-buf,y+roi+buf,width) # +/- 20 pixels in vertical direction, -20 to the right edge if x3<0: x3 = 0 if y3<0: y3 = 0 if x4>=width: x4 = width-1 if y4>height: y4 = height #cv.cvSetImageROI(rightimage, (y3,x3,y4,x4)) rightsub_rect = cv.cvRect(x3,y3,(x4-x3),(y4-y3)) rightsub = cv.cvGetSubRect(rightimage, rightsub_rect) # result matrix should be (W - w + 1) x (H - h + 1) where WxH are template dimensions, wxh are rightsub dimensions W = x4-x3 H = y4-y3 w = x2-x1 h = y2-y1 resy = (y4-y3)-(y2-y1)+1 resx = (x4-x3)-(x2-x1)+1 resultmat = cv.cvCreateImage((resx, resy), 32, 1) cv.cvZero(resultmat) # match template image in a subportion of rightimage cv.cvMatchTemplate(rightsub, template, resultmat, cv.CV_TM_SQDIFF) min_val, max_val, min_point, max_point = cv.cvMinMaxLoc(resultmat) cv.cvNormalize(resultmat, resultmat, 1, 0, cv.CV_MINMAX) depth = stereo.depth(x, x3+min_point.x, max_pixels=width/2) if debug: print "Input image: %ix%i, target: (%i,%i)" % (width,height,x,y) print "Template box: (%i,%i) to (%i,%i)" % (x1, y1, x2, y2) print "Search area: (%i,%i) to (%i,%i)" % (x3, y3, x4, y4) print "%ix%i, %ix%i" % (W,H,w,h) print "Result matrix %ix%i" % (resx, resy) print "stereo.depth(%i,%i,max_pixels=%i)" % (x, min_point.x+x3,width/2) if depth[0]: print "Depth: ", depth[0], "(cm)" #cv.cvRectangle(rightimage, cv.cvPoint(x1,y1), cv.cvPoint(x2,y2), (255,0,0)) cv.cvRectangle(rightimage, cv.cvPoint(min_point.x+x3,min_point.y+y3), cv.cvPoint(min_point.x+x3+roi*2,min_point.y+y3+roi*2), (0,255,0)) cv.cvRectangle(rightimage, cv.cvPoint(x3,y3), cv.cvPoint(x4,y4), (0,0,255)) cv.cvRectangle(leftimage, cv.cvPoint(x1,y1), cv.cvPoint(x2,y2), (255,0,0)) #cv.cvRectangle(leftimage, cv.cvPoint(min_point.x+x3,min_point.y+y3), cv.cvPoint(min_point.x+x3+roi*2,min_point.y+y3+roi*2), (0,255,0)) cv.cvRectangle(leftimage, cv.cvPoint(x3,y3), cv.cvPoint(x4,y4), (0,0,255)) if depth[0]: cv.cvPutText(leftimage, "%5f(cm)" % depth[0], (x1,y1), font, (255,255,255)) highgui.cvShowImage("depthmatch - template", template) highgui.cvShowImage("depthmatch - match", resultmat) highgui.cvShowImage("depthmatch - right", rightimage) highgui.cvShowImage("depthmatch - left", leftimage)
def compute_saliency(image): global thresh global scale saliency_scale = int(math.pow(2,scale)); bw_im1 = cv.cvCreateImage(cv.cvGetSize(image), cv.IPL_DEPTH_8U,1) cv.cvCvtColor(image, bw_im1, cv.CV_BGR2GRAY) bw_im = cv.cvCreateImage(cv.cvSize(saliency_scale,saliency_scale), cv.IPL_DEPTH_8U,1) cv.cvResize(bw_im1, bw_im) highgui.cvShowImage("BW", bw_im) realInput = cv.cvCreateImage( cv.cvGetSize(bw_im), cv.IPL_DEPTH_32F, 1); imaginaryInput = cv.cvCreateImage( cv.cvGetSize(bw_im), cv.IPL_DEPTH_32F, 1); complexInput = cv.cvCreateImage( cv.cvGetSize(bw_im), cv.IPL_DEPTH_32F, 2); cv.cvScale(bw_im, realInput, 1.0, 0.0); cv.cvZero(imaginaryInput); cv.cvMerge(realInput, imaginaryInput, None, None, complexInput); dft_M = saliency_scale #cv.cvGetOptimalDFTSize( bw_im.height - 1 ); dft_N = saliency_scale #cv.cvGetOptimalDFTSize( bw_im.width - 1 ); dft_A = cv.cvCreateMat( dft_M, dft_N, cv.CV_32FC2 ); image_Re = cv.cvCreateImage( cv.cvSize(dft_N, dft_M), cv.IPL_DEPTH_32F, 1); image_Im = cv.cvCreateImage( cv.cvSize(dft_N, dft_M), cv.IPL_DEPTH_32F, 1); # copy A to dft_A and pad dft_A with zeros tmp = cv.cvGetSubRect( dft_A, cv.cvRect(0,0, bw_im.width, bw_im.height)); cv.cvCopy( complexInput, tmp, None ); if(dft_A.width > bw_im.width): tmp = cv.cvGetSubRect( dft_A, cv.cvRect(bw_im.width,0, dft_N - bw_im.width, bw_im.height)); cv.cvZero( tmp ); cv.cvDFT( dft_A, dft_A, cv.CV_DXT_FORWARD, complexInput.height ); cv.cvSplit( dft_A, image_Re, image_Im, None, None ); # Compute the phase angle image_Mag = cv.cvCreateImage(cv.cvSize(dft_N, dft_M), cv.IPL_DEPTH_32F, 1); image_Phase = cv.cvCreateImage(cv.cvSize(dft_N, dft_M), cv.IPL_DEPTH_32F, 1); #compute the phase of the spectrum cv.cvCartToPolar(image_Re, image_Im, image_Mag, image_Phase, 0) log_mag = cv.cvCreateImage(cv.cvSize(dft_N, dft_M), cv.IPL_DEPTH_32F, 1); cv.cvLog(image_Mag, log_mag) #Box filter the magnitude, then take the difference image_Mag_Filt = cv.cvCreateImage(cv.cvSize(dft_N, dft_M), cv.IPL_DEPTH_32F, 1); filt = cv.cvCreateMat(3,3, cv.CV_32FC1); cv.cvSet(filt,cv.cvScalarAll(-1.0/9.0)) cv.cvFilter2D(log_mag, image_Mag_Filt, filt, cv.cvPoint(-1,-1)) cv.cvAdd(log_mag, image_Mag_Filt, log_mag, None) cv.cvExp(log_mag, log_mag) cv.cvPolarToCart(log_mag, image_Phase, image_Re, image_Im,0); cv.cvMerge(image_Re, image_Im, None, None, dft_A) cv.cvDFT( dft_A, dft_A, cv.CV_DXT_INVERSE, complexInput.height) tmp = cv.cvGetSubRect( dft_A, cv.cvRect(0,0, bw_im.width, bw_im.height)); cv.cvCopy( tmp, complexInput, None ); cv.cvSplit(complexInput, realInput, imaginaryInput, None, None) min, max = cv.cvMinMaxLoc(realInput); #cv.cvScale(realInput, realInput, 1.0/(max-min), 1.0*(-min)/(max-min)); cv.cvSmooth(realInput, realInput); threshold = thresh/100.0*cv.cvAvg(realInput)[0] cv.cvThreshold(realInput, realInput, threshold, 1.0, cv.CV_THRESH_BINARY) tmp_img = cv.cvCreateImage(cv.cvGetSize(bw_im1),cv.IPL_DEPTH_32F, 1) cv.cvResize(realInput,tmp_img) cv.cvScale(tmp_img, bw_im1, 255,0) return bw_im1
def draw(self,screen) : if not self.is_calibrated : screen.fill(THECOLORS["white"]) checkerboard = pygame.image.load('checkerboard.png') checkerboard = pygame.transform.scale(checkerboard,self.dims) screen.blit(checkerboard,(0,0)) pygame.display.flip() corners = self.lazer._calibrate_camera() self.is_calibrated = True else : #frame = self.lazer.get_curr_frame() #screen.blit(frame,(0,0)) pass """ pt = self.input.get_com() if pt is None : pt = (0,0) if self.blink_phase == 0 : self.pointer_sprite.update(pt) else : self.blink_phase -= 1 self.pointer_sprite.update(None) if self.pointer_sprite.on : self.curr_line.append(pt) else : self.curr_line = [] self.lines.append(self.curr_line) if self.pointer_sprite.clicked : self.button.click() for l in self.lines : if len(l) > 1 : pygame.draw.aalines(screen,THECOLORS["black"],False,l,1) """ # don't try this at home, kids frame = self.lazer._get_cv_frame() thresh = 248 red = self.lazer._get_color_frame('r',frame) cvpt_min = cv.cvPoint(0,0) cvpt_max = cv.cvPoint(0,0) t = cv.cvMinMaxLoc(red,cvpt_min,cvpt_max) #print t,cvpt_min,cvpt_max #red = self.lazer._get_scaled_frame(red) #red = self.lazer._get_threshold_frame(frame=red,thresh=thresh,type=CV_THRESH_BINARY) blue = self.lazer._get_color_frame('b',frame) #blue = self.lazer._get_threshold_frame(frame=blue,thresh=thresh,type=CV_THRESH_BINARY) blue = self.lazer._cv_to_pygame(blue,channel=2) blue = pygame.transform.scale(blue,(self.dims[0]/2,self.dims[1]/2)) green = self.lazer._get_color_frame('g',frame) #green = self.lazer._get_threshold_frame(frame=green,thresh=thresh,type=CV_THRESH_BINARY) green = self.lazer._cv_to_pygame(green,channel=1) green = pygame.transform.scale(green,(self.dims[0]/2,self.dims[1]/2)) value = self.lazer._get_hsv_frame(frame) value = self.lazer._get_frame_channel(value,2) value = self.lazer._get_shifted_frame(-215,value) #value = self.lazer._get_threshold_frame(frame=value,thresh=thresh,type=CV_THRESH_BINARY) comb = self.lazer._multiply_frames(red,value) comb = self.lazer._cv_to_pygame(comb,channel=5) comb = pygame.transform.scale(comb,(self.dims[0]/2,self.dims[1]/2)) red = self.lazer._cv_to_pygame(red,channel=0) red = pygame.transform.scale(red,(self.dims[0]/2,self.dims[1]/2)) value = self.lazer._cv_to_pygame(value,channel=-1) value = pygame.transform.scale(value,(self.dims[0]/2,self.dims[1]/2)) #frame = self.lazer.get_curr_frame() frame = self.lazer._cv_to_pygame(frame) frame = pygame.transform.scale(frame,(self.dims[0]/2,self.dims[1]/2)) screen.blit(red,(0,0)) screen.blit(blue,(self.dims[0]/2,0)) screen.blit(green,(0,self.dims[1]/2)) screen.blit(frame,(self.dims[0]/2,self.dims[1]/2)) screen.fill((100,100,100,255)) if t[1] >= 0 : pygame.draw.circle(screen,THECOLORS["red"],(cvpt_max.x,cvpt_max.y),5,1) pygame.draw.circle(screen,THECOLORS["white"],(cvpt_max.x,cvpt_max.y),3,0)