def compute_feature( reference, frame, radius = 3 ):
density = 0.0
feature = np.array( [0.] * FEATURE_LENGTH, dtype = np.float32 )
sum_ang = 0.0
sum_mag = 0.0
for i in xrange( len( frame ) ):
dist = momo.distance( frame[i][:2], reference[:2] )
if dist < radius:
density += 1
rel_v = frame[i][2:] - reference[2:]
rel_x = frame[i][:2] - reference[:2]
l_v = np.linalg.norm( rel_v )
l_x = np.linalg.norm( rel_x )
a = np.dot( rel_v / l_v, rel_x / l_x )
sum_ang += a
sum_mag += l_v
if density > 0:
feature[dns_i + max_idx( density, DENSITIES )] = 1
speed = sum_mag / density
feature[spd_i + max_idx( speed, SPEEDS )] = 1
angle = sum_ang / density
feature[dir_i + max_idx( angle, ANGLES )] = 1
feature[FEATURE_LENGTH - 1] = 1
return feature
def compute_feature(reference, frame, radius=3):
density = 0
feature = np.array([0.0] * FEATURE_LENGTH, dtype=np.float32)
bin_count = np.zeros((3,), dtype=np.float32)
bin_sum = np.zeros((3,), dtype=np.float32)
for i in xrange(len(frame)):
dist = momo.distance(frame[i][:2], reference[:2])
if dist < radius:
density += 1
rel_v = frame[i][2:] - reference[2:]
rel_x = frame[i][:2] - reference[:2]
l_v = np.linalg.norm(rel_v)
l_x = np.linalg.norm(rel_x)
a = np.dot(rel_v / l_v, rel_x / l_x)
i = max_idx(a, ANGLES)
bin_count[i] += 1
bin_sum[i] += l_v
if density > 0:
feature[max_idx(density, DENSITIES)] = 1
for angle in xrange(3):
if bin_count[angle] > 0:
l = bin_sum[angle] / bin_count[angle]
feature[3 + angle * 3 + max_idx(l, SPEEDS)] = 1
feature[FEATURE_LENGTH - 1] = 1
return feature
def compute_feature( reference, frame, radius = 3 ):
density = 0
avg_velocity = np.array( [0.] * 2 )
feature = np.array( [0.] * FEATURE_LENGTH, dtype = np.float32 )
for i in xrange( len( frame ) ):
dist = momo.distance( frame[i][:2], reference[:2] )
if dist < radius:
density += 1
avg_velocity += frame[i][2:]
if ( density >= 3 ):
feature[3] = 1
else:
feature[density] = 1
if density == 0:
feature[16] = 1
else:
avg_velocity /= density
avg_velocity = reference[2:] - avg_velocity
avg_velocity[1] = abs( avg_velocity[1] )
angle = avg_velocity / np.linalg.norm( avg_velocity )
angle_index = 0
max_dot = -1
for i in xrange( 4 ):
a = np.dot( ANGLES[i], angle )
if a > max_dot:
max_dot = a
angle_index = i
speed = np.linalg.norm( avg_velocity )
speed_index = 0;
for i in xrange( 3 ):
if speed >= SPEEDS[i]:
speed_index = i
feature[4 + angle_index * 3 + speed_index] = 1
feature[17] = 1
return feature