def associate_detections_BS(_detections, _predictions):
_assignments = Hungarian_method(_detections, _predictions)
_missing = []
_insects = [i[0] for i in _predictions]
_not_associated = np.zeros(shape=(0, 3))
for _nass in (_assignments[len(_insects):]):
_not_associated = np.vstack([_not_associated, (_detections[_nass])])
_associations_BS = np.zeros(shape=(0, 4))
for ass in np.arange(len(_insects)):
_record = _assignments[ass]
if (_record <= len(_detections) - 1):
_xc, _yc, _area = _detections[_assignments[ass]][0], _detections[
_assignments[ass]][1], _detections[_assignments[ass]][2]
_dist = cal_dist(_xc, _yc, _predictions[ass][1],
_predictions[ass][2])
if (_dist > max_dist_bs):
_missing.append(_predictions[ass][0])
else:
_associations_BS = np.vstack([
_associations_BS,
(int(float(_predictions[ass][0])), int(_xc), int(_yc),
int(_area))
])
else:
_missing.append(_predictions[ass][0])
return _associations_BS, _missing, _not_associated
def cal_cum_distance(_last_detections):
cum_distance = 0
for _d in np.arange(0, len(_last_detections) - 1, 1):
cum_distance += cal_dist(_last_detections[_d][0],
_last_detections[_d][1],
_last_detections[_d + 1][0],
_last_detections[_d + 1][1])
return cum_distance
def Hungarian_method(_detections, _predictions):
num_detections, num_predictions = len(_detections), len(_predictions)
mat_shape = max(num_detections, num_predictions)
hun_matrix = np.full((mat_shape, mat_shape), 0)
for p in np.arange(num_predictions):
for d in np.arange(num_detections):
hun_matrix[p][d] = cal_dist(_predictions[p][1], _predictions[p][2],
_detections[d][0], _detections[d][1])
row_ind, col_ind = linear_sum_assignment(hun_matrix)
return col_ind
def verify_new_insect(_pot_new_insect, _associated_det_BS):
recorded_BS = []
for bsdet in _associated_det_BS:
for pni in np.arange(len(_pot_new_insect)):
dist = cal_dist(bsdet[1], bsdet[2],
int(float(_pot_new_insect[pni][0])),
int(float(_pot_new_insect[pni][1])))
if dist <= max_dist_bs:
recorded_BS.append(pni)
_pot_new_insect = np.delete(_pot_new_insect, recorded_BS, 0)
return _pot_new_insect
def cal_CoG_DL(result):
_x_DL, _y_DL, _body_area, _radius = 0, 0, 0, 0
_x_TL = int(float(result[0]))
_y_TL = int(float(result[1]))
_x_BR = int(float(result[2]))
_y_BR = int(float(result[3]))
_x_DL = int(round((_x_TL + _x_BR) / 2))
_y_DL = int(round((_y_TL + _y_BR) / 2))
_radius = cal_dist(_x_TL, _y_TL, _x_DL, _y_DL)
_body_area = cal_bodyArea_DL(_x_TL, _y_TL, _x_BR, _y_BR)
return _x_DL, _y_DL, _body_area, _radius
def check_on_flower(_prediction, _on_flower=False):
_x = _prediction[1]
_y = _prediction[2]
_flowers = get_flower_details()
for flower in _flowers:
dist_from_c = cal_dist(_x, _y, flower[0], flower[1])
if dist_from_c <= flower[2]:
_on_flower = True
break
else:
pass
return _on_flower
def check_sight_coordinates(_record, outside_spot=True):
if pt_cfg.POLYTRACK.DL_DARK_SPOTS:
_x = int(float(_record[0]))
_y = int(float(_record[1]))
dark_spots = pt_cfg.POLYTRACK.RECORDED_DARK_SPOTS
spot_radius = int(pt_cfg.POLYTRACK.DL_DARK_SPOTS_RADIUS)
for spot in dark_spots:
spot_x = spot[0]
spot_y = spot[1]
if (cal_dist(_x, _y, spot_x, spot_y) <= spot_radius):
outside_spot = False
break
else:
pass
else:
outside_spot = True
return outside_spot
def associate_detections_DL(_detections, _predictions):
_missing = []
_assignments = Hungarian_method(_detections, _predictions)
_insects = [i[0] for i in _predictions]
_not_associated = np.zeros(shape=(0, 5))
for _nass in (_assignments[len(_insects):]):
_not_associated = np.vstack([_not_associated, (_detections[_nass])])
_associations_DL = np.zeros(shape=(0, 6))
for ass in np.arange(len(_insects)):
_record = _assignments[ass]
if (_record <= len(_detections) - 1):
_xc, _yc, _area, _lable, _conf = _detections[_assignments[ass]][
0], _detections[_assignments[ass]][1], _detections[
_assignments[ass]][2], _detections[_assignments[ass]][
3], _detections[_assignments[ass]][4]
_dist = cal_dist(_xc, _yc, _predictions[ass][1],
_predictions[ass][2])
if (_dist > max_dist_dl):
_missing.append(_predictions[ass][0])
else:
_associations_DL = np.vstack([
_associations_DL,
(_predictions[ass][0], _detections[_assignments[ass]][0],
_detections[_assignments[ass]][1],
_detections[_assignments[ass]][2],
_detections[_assignments[ass]][3],
_detections[_assignments[ass]][4])
])
else:
_missing.append(_predictions[ass][0])
return _associations_DL, _missing, _not_associated