def dvs_img(cloud, shape, K, D):
cmb = pydvs.dvs_img(cloud, shape, K=K, D=D)
cmb[:, :, 0] *= global_scale_pp
cmb[:, :, 1] *= 255.0 / slice_width
cmb[:, :, 2] *= global_scale_pn
return cmb
def dvs_img(cloud, shape, K, D, slice_width, mode=0):
cmb = pydvs.dvs_img(cloud, shape, K=K, D=D)
ncnt = cmb[:, :, 0]
time = cmb[:, :, 1]
pcnt = cmb[:, :, 2]
cnt = pcnt + ncnt
# Scale up to be able to save as uint8
# For visualization only. May cause overflow if slice_width is large
cmb[:, :, 0] *= 50
cmb[:, :, 1] *= 255.0 / slice_width
cmb[:, :, 2] *= 50
if (mode == 1):
cmb = np.dstack((time, pcnt, ncnt))
return cmb
def __init__(self, shape, cloud):
random.seed()
self.shape = shape
self.width = 0
self.dgrad = np.zeros((self.shape[0], self.shape[1], 3))
if (cloud.shape[0] > 0):
self.width = cloud[-1][0] - cloud[0][0]
self.dvs_img = pydvs.dvs_img(cloud,
self.shape,
model=[0, 0, 0, 0],
scale=1,
K=None,
D=None)
#self.dvs_img[:,:,1] = cv2.medianBlur(self.dvs_img[:,:,1], 5)
self.dvs_img[:, :, 1] = cv2.GaussianBlur(self.dvs_img[:, :, 1], (7, 7),
0)
dgrad = np.zeros((self.dvs_img.shape[0], self.dvs_img.shape[1], 2),
dtype=np.float32)
self.x_err2, self.y_err2, self.yaw_err, self.z_err2, self.e_count, self.nz_avg = \
pydvs.dvs_flow_err(self.dvs_img, dgrad)
dgrad = np.zeros((self.dvs_img.shape[0], self.dvs_img.shape[1], 2),
dtype=np.float32)
self.x_err, self.y_err, self.yaw_err, self.z_err, self.e_count, self.nz_avg = \
pydvs.dvs_err(self.dvs_img, dgrad)
self.x_err *= 10
self.y_err *= 10
self.z_err /= 5
self.x_err2 /= 5
self.y_err2 /= 5
self.z_err2 /= 500
self.e_count /= 6250
self.e_count -= 1
self.nz_avg /= 4
self.nz_avg -= 1
self.p_count = np.sum(self.dvs_img[:, :, 0]) / 20000
self.n_count = np.sum(self.dvs_img[:, :, 2]) / 30000
self.g_count = self.p_count + self.n_count / 50000
self.p_count -= 1
self.n_count -= 1
self.g_count -= 1
#function image2vec
#Important function! first one really related to HBV vectors
dgrad[:, :, 0] = np.sqrt((dgrad[:, :, 0] * dgrad[:, :, 0]) +
(dgrad[:, :, 1] * dgrad[:, :, 1])) * np.sign(
dgrad[:, :, 0] * dgrad[:, :, 1])
dgrad[:, :, 1] = dgrad[:, :, 0]
dgrad[dgrad > 1.8] = 1.8
self.dgrad[:, :, 0] = np.copy(dgrad[:, :, 0])
self.dgrad[:, :, 1] = np.copy(dgrad[:, :, 0])
self.dgrad[:, :, 2] = np.copy(dgrad[:, :, 0])
self.vec, self.feature_list, self.rescaled_feat = self.image2vec(dgrad)
return
#print (self.x_err, self.y_err, self.z_err, self.yaw_err, self.e_count, self.nz_avg)
# Visualization
c_img = self.dvs_img[:, :, 0] + self.dvs_img[:, :, 2]
c_img = np.dstack((c_img, c_img, c_img)) * 0.5 / (self.nz_avg + 1e-3)
self.dvs_img[:, :, 1] *= 1.0 / self.width
t_img = np.dstack(
(self.dvs_img[:, :, 1], self.dvs_img[:, :, 1], self.dvs_img[:, :,
1]))
G_img = colorizeImage(dgrad[:, :, 0], dgrad[:, :, 1])
self.vis = np.hstack((t_img, G_img))
cv2.namedWindow('GUI', cv2.WINDOW_NORMAL)
cv2.imshow('GUI', np.hstack((t_img, G_img)))
cv2.waitKey(0)
def __init__(self, shape, cloud):
random.seed()
self.shape = shape
self.width = 0
if (cloud.shape[0] > 0):
self.width = cloud[-1][0] - cloud[0][0]
# Compute images according to the model
self.dvs_img = pydvs.dvs_img(cloud, self.shape, model=[0, 0, 0, 0],
scale=1, K=None, D=None)
#self.dvs_img[:,:,1] = cv2.medianBlur(self.dvs_img[:,:,1], 5)
self.dvs_img[:,:,1] = cv2.GaussianBlur(self.dvs_img[:,:,1], (7,7), 0)
#dvs_img = np.copy(dvs_img[:50,:50,:])
# Compute errors on the images
dgrad = np.zeros((self.dvs_img.shape[0], self.dvs_img.shape[1], 2), dtype=np.float32)
self.x_err2, self.y_err2, self.yaw_err, self.z_err2, self.e_count, self.nz_avg = \
pydvs.dvs_flow_err(self.dvs_img, dgrad)
dgrad = np.zeros((self.dvs_img.shape[0], self.dvs_img.shape[1], 2), dtype=np.float32)
self.x_err, self.y_err, self.yaw_err, self.z_err, self.e_count, self.nz_avg = \
pydvs.dvs_err(self.dvs_img, dgrad)
self.x_err *= 10
self.y_err *= 10
self.z_err /= 5
self.x_err2 /= 5
self.y_err2 /= 5
self.z_err2 /= 500
self.e_count /= 6250
self.e_count -= 1
self.nz_avg /= 4
self.nz_avg -= 1
self.p_count = np.sum(self.dvs_img[:,:,0]) / 20000
self.n_count = np.sum(self.dvs_img[:,:,2]) / 30000
self.g_count = self.p_count + self.n_count / 50000
self.p_count -= 1
self.n_count -= 1
self.g_count -= 1
self.vec = self.image2vec(dgrad)
return
#print (self.x_err, self.y_err, self.z_err, self.yaw_err, self.e_count, self.nz_avg)
# Visualization
c_img = self.dvs_img[:,:,0] + self.dvs_img[:,:,2]
c_img = np.dstack((c_img, c_img, c_img)) * 0.5 / (self.nz_avg + 1e-3)
self.dvs_img[:,:,1] *= 1.0 / self.width
t_img = np.dstack((self.dvs_img[:,:,1], self.dvs_img[:,:,1], self.dvs_img[:,:,1]))
G_img = colorizeImage(dgrad[:,:,0], dgrad[:,:,1])
self.vis = np.hstack((t_img, G_img))
cv2.namedWindow('GUI', cv2.WINDOW_NORMAL)
cv2.imshow('GUI', np.hstack((t_img, G_img)))
cv2.waitKey(0)