Go to the documentation of this file.00001
00002
00003
00004
00005
00006
00007
00008
00009
00010
00011
00012
00013
00014
00015
00016
00017
00018
00019
00020
00021
00022
00023
00024
00025
00026
00027
00028
00029
00030
00031
00032
00033
00034
00035
00036
00037
00038 from __future__ import division
00039 import roslib
00040 roslib.load_manifest('object_manipulator')
00041 import rospy
00042 import scipy
00043 import math
00044 from sensor_msgs.msg import RegionOfInterest
00045 import pdb
00046
00047
00048
00049 def compute_cluster_mask(camera_info, cluster, dist = 1):
00050 indices = []
00051
00052
00053
00054 for point in cluster.points:
00055 (x,y) = compute_image_point(camera_info, point)
00056 for i in range(-dist,dist+1):
00057 if x+i < 0 or x+i > camera_info.width:
00058 continue
00059 for j in range(-dist,dist+1):
00060 if y+j < 0 or y+j > camera_info.height:
00061 continue
00062 index = image_points_to_indices(camera_info.width, [x+i], [y+j])[0]
00063 indices.append(index)
00064
00065
00066 unique_indices = set(indices)
00067 return list(unique_indices)
00068
00069
00070
00071 def image_points_to_indices(width, xvals, yvals):
00072 return [y*width + x for (x,y) in zip(xvals, yvals)]
00073
00074
00075
00076 def indices_to_image_points(width, indices):
00077 xvals = [ind - width*(math.floor(ind / width)) for ind in indices]
00078 yvals = [math.floor(ind / width) for ind in indices]
00079 return (xvals, yvals)
00080
00081
00082
00083 def compute_image_point(camera_info, point):
00084
00085 trans_mat = scipy.matrix(scipy.identity(4))
00086 for i in range(3):
00087 for j in range(4):
00088 trans_mat[i,j] = camera_info.P[i*4+j]
00089 point4 = scipy.matrix(scipy.ones([4,1]))
00090 point4[0] = point.x
00091 point4[1] = point.y
00092 point4[2] = point.z
00093
00094 image_pt = trans_mat * point4
00095 x = int(math.floor(image_pt[0,0] / image_pt[2,0]))
00096 y = int(math.floor(image_pt[1,0] / image_pt[2,0]))
00097
00098 return (x,y)
00099
00100
00101
00102 def generate_rect_mask(x_offset, y_offset, height, width, row_step):
00103
00104 point_indices = []
00105 for y in range(height+1):
00106 row = range(x_offset+y*row_step, x_offset+y*row_step+width+1)
00107 point_indices.extend(row)
00108 return point_indices
00109
00110
00111
00112 def compute_roi_from_indices(indices, width, height, padding = 30):
00113
00114 xvals = [ind - width*(math.floor(ind / width)) for ind in indices]
00115 yvals = [math.floor(ind / width) for ind in indices]
00116 roi = RegionOfInterest()
00117 roi.x_offset = max(0, min(xvals) - padding)
00118 roi.y_offset = max(0, min(yvals) - padding)
00119
00120 roi.height = max(yvals) - roi.y_offset + padding
00121 if roi.height + roi.y_offset > height:
00122 roi.height = height - roi.y_offset
00123 roi.width = max(xvals) - roi.x_offset + padding
00124 if roi.width + roi.x_offset > width:
00125 roi.width = width - roi.x_offset
00126 roi.do_rectify = 0
00127
00128 return roi