diff --git a/scripts/Depth2Disparity/depth_to_disparity.py b/scripts/Depth2Disparity/depth_to_disparity.py
index d13d3dc369c7f39d170123eedc2370a53092e195..10c0df9b5972fba4d15013290d3a3bdbad7beaca 100644
--- a/scripts/Depth2Disparity/depth_to_disparity.py
+++ b/scripts/Depth2Disparity/depth_to_disparity.py
@@ -109,73 +109,113 @@ class Depth2Disparity:
abs_depth_map = a * rel_depth_map + b
print("Applied transformation to entire depth map")
- abs_depth_map = np.maximum(abs_depth_map, 0.01)
+ #abs_depth_map = np.maximum(abs_depth_map, 0.01)
- #plt.imshow(abs_depth_map, cmap='gray')
- #plt.colorbar()
- #plt.show()
+ plt.imshow(abs_depth_map, cmap='gray')
+ plt.colorbar()
+ plt.show()
# this should not be normalised, the values in the array should equate to literal distances
return abs_depth_map
def _depth2disparity(self, abs_depth_map):
-
- # Get image dimensions
- # Get image dimensions
- height, width = abs_depth_map.shape
-
- # Calculate step sizes
- unit_w = 2.0 / width # Longitude step size
- unit_h = 1.0 / height # Latitude step size
-
- # Initialize disparity map
- disparity_map = np.zeros_like(abs_depth_map, dtype=np.float32)
-
- # Iterate over each pixel
- for i in range(height):
- # Calculate latitude angle for the row
- theta_t = i * unit_h * np.pi
-
- for j in range(width):
- # Retrieve the absolute depth (r_t) for this pixel
- r_t = abs_depth_map[i, j]
-
- # Avoid invalid or infinite depth values
- if r_t <= 0:
- disparity_map[i, j] = 0
- continue
-
- try:
- # Compute denominator for the formula
- epsilon = 1e-8 # Small value to prevent instability
- tan_theta_t = np.tan(theta_t) if np.abs(np.cos(theta_t)) > epsilon else np.sign(np.sin(theta_t)) * np.inf
- denominator = r_t * np.sin(theta_t) + r_t * np.cos(theta_t) * tan_theta_t
-
- # Avoid invalid denominator values
- if denominator <= 0:
- disparity_map[i, j] = 0
- continue
-
- # Calculate angular disparity
- angle_disp = np.arctan(self.baseline / denominator) - theta_t
-
- # Convert angular disparity to pixel disparity
- pixel_disp = (angle_disp / (unit_h * np.pi) - self.disp_offset) * self.disp_scale
- disparity_map[i, j] = pixel_disp
-
- except ZeroDivisionError:
- disparity_map[i, j] = 0
-
- plt.imshow(disparity_map, cmap='gray')
- plt.colorbar()
- plt.show()
-
- # Normalize disparity map for visualization
- disparity_map_normalized = (disparity_map - disparity_map.min()) / (disparity_map.max() - disparity_map.min())
- disparity_map_normalized *= 255
- disparity_map_normalized = disparity_map_normalized.astype(np.uint8)
-
- return disparity_map, disparity_map_normalized
+ """
+ Convert absolute depth map to disparity map for 360-degree images.
+
+ Parameters:
+ abs_depth_map (numpy.ndarray): Absolute depth map with pixel values as distances in meters.
+
+ Returns:
+ disparity_map (numpy.ndarray): Disparity map with values representing disparity.
+ disparity_map_normalized (numpy.ndarray): Normalized disparity map for visualization.
+ """
+ # Define parameters for the conversion
+ height, width = abs_depth_map.shape
+
+ # Latitude grid (θ_l)
+ epsilon = 1e-6
+ latitudes = np.linspace(-np.pi / 2, np.pi / 2, height)
+ latitudes_grid = np.tile(latitudes[:, np.newaxis], (1, width))
+
+ # Clamp extreme latitudes to avoid instability
+ latitudes_grid = np.clip(latitudes_grid, -np.pi/2 + epsilon, np.pi/2 - epsilon)
+
+ plt.imshow(latitudes_grid, cmap="jet")
+ plt.colorbar()
+ plt.title("Latitude Grid")
+ plt.show()
+
+ self.baseline = 0.176
+
+
+ # Calculate angular disparity Δθ(x, y)
+ # Δθ(x, y) = arctan(B / (r_t * sin(θ_l) + r_t * cos(θ_l) * tan(θ_l))) - θ_l
+ epsilon = 1e-6 # Small value to prevent instability
+ denominator = (
+ abs_depth_map * np.sin(latitudes_grid) +
+ abs_depth_map * np.cos(latitudes_grid) * np.tan(latitudes_grid) +
+ epsilon
+ )
+ angular_disparity = np.arctan(self.baseline / denominator) - latitudes_grid
+
+ self.disp_scale = 10
+ self.disp_offset = np.median(angular_disparity)
+
+ # Normalize the disparity map for saving
+ pixel_angle_scale = 1 # Assume scale factor is 1 if not provided (can be adjusted if needed)
+ disparity_map = (angular_disparity / pixel_angle_scale - self.disp_offset) * self.disp_scale
+
+ # Normalize for visualization
+ disparity_map_normalized = (disparity_map - disparity_map.min()) / (disparity_map.max() - disparity_map.min())
+ disparity_map_normalized *= 255
+ disparity_map_normalized = disparity_map_normalized.astype(np.uint8)
+
+ # Debugging visualization
+ plt.imshow(disparity_map, cmap="jet")
+ plt.colorbar()
+ plt.title("Disparity Map (Debugging)")
+ plt.show()
+
+ return disparity_map, disparity_map_normalized
+
+ def __depth2disparity(self, abs_depth_map):
+ height, width = abs_depth_map.shape
+ unit_w = 2.0 / width # Horizontal unit angle
+ unit_h = 1.0 / height # Vertical unit angle
+ epsilon = 1e-8 # Small value to prevent instability
+
+ # Latitude angles (theta_t) for each row
+ theta_t_grid = np.linspace(0, np.pi, height).reshape(height, 1)
+
+ # Ensure no instability in tangent calculation
+ tan_theta_t = np.tan(np.clip(theta_t_grid, -np.pi/2 + epsilon, np.pi/2 - epsilon))
+ # Broadcast depth map and latitude grid
+ r_t = abs_depth_map # Absolute depth values (meters)
+ sin_theta_t = np.sin(theta_t_grid)
+ cos_theta_t = np.cos(theta_t_grid)
+
+ # Compute denominator
+ denominator = r_t * sin_theta_t + r_t * cos_theta_t * tan_theta_t + epsilon
+
+ # Avoid invalid denominator values
+ denominator[denominator <= 0] = epsilon
+
+ # Calculate angular disparity
+ angle_disp = np.arctan(self.baseline / denominator) - theta_t_grid
+
+ # Convert angular disparity to pixel disparity
+ disparity_map = (angle_disp / (unit_h * np.pi) - self.disp_offset) * self.disp_scale
+
+ plt.imshow(disparity_map, cmap='gray')
+ plt.colorbar()
+ plt.show()
+
+ # Normalize disparity map for visualization
+ disparity_map_normalized = (disparity_map - disparity_map.min()) / (disparity_map.max() - disparity_map.min())
+ disparity_map_normalized *= 255
+ disparity_map_normalized = disparity_map_normalized.astype(np.uint8)
+
+ return disparity_map, disparity_map_normalized
\ No newline at end of file