Add avoidance computation

Author: schroedtert

pedpy/__init__.py

@@ -29,6 +29,11 @@
     compute_voronoi_density,
 )
 from .methods.flow_calculator import compute_flow, compute_n_t
+from .methods.foo_calculator import (
+    IntrusionMethod,
+    compute_avoidance,
+    compute_intrusion,
+)
 from .methods.method_utils import (
     Cutoff,
     compute_frame_range_in_area,
@@ -55,7 +60,6 @@
     compute_passing_speed,
     compute_voronoi_speed,
 )
-from .methods.foo_calculator import compute_intrusion, IntrusionMethod
 from .plotting.plotting import (
     PEDPY_BLUE,
     PEDPY_GREEN,
@@ -119,6 +123,7 @@
     "compute_voronoi_speed",
     "compute_intrusion",
     "IntrusionMethod",
+    "compute_avoidance",
     "PEDPY_BLUE",
     "PEDPY_GREEN",
     "PEDPY_GREY",

pedpy/column_identifier.py

@@ -31,3 +31,4 @@
 WINDOW_SIZE_COL: Final = "window_size"
 
 INTRUSION_COL: Final = "intrusion"
+AVOIDANCE_COL: Final = "avoidance"

pedpy/methods/foo_calculator.py

@@ -3,10 +3,22 @@
 
 from enum import Enum
 
-from pedpy.data.trajectory_data import TrajectoryData
-from .method_utils import _compute_individual_distances
-from pedpy.column_identifier import FRAME_COL, ID_COL, INTRUSION_COL
+import numpy as np
 import pandas
+import shapely
+
+from pedpy.column_identifier import (
+    AVOIDANCE_COL,
+    FRAME_COL,
+    ID_COL,
+    INTRUSION_COL,
+)
+from pedpy.data.trajectory_data import TrajectoryData
+from pedpy.methods.method_utils import (
+    SpeedCalculation,
+    _compute_individual_distances,
+)
+from pedpy.methods.speed_calculator import compute_individual_speed
 
 
 class IntrusionMethod(Enum):  # pylint: disable=too-few-public-methods
@@ -50,3 +62,93 @@ def compute_intrusion(
     )
 
     return intrusion
+
+
+def compute_avoidance(
+    *,
+    traj_data: TrajectoryData,
+    frame_step: int,
+    radius: float = 0.2,
+    tau_0: float,
+) -> pandas.DataFrame:
+    """_summary_.
+
+    TODO add documentation here
+
+    Args:
+        traj_data (TrajectoryData): _description_
+        frame_step (int): _description_
+        radius (float): _description_
+        tau_0 (float): _description_
+
+    Returns:
+        pandas.DataFrame: _description_
+    """
+    velocity = compute_individual_speed(
+        traj_data=traj_data,
+        frame_step=frame_step,
+        compute_velocity=True,
+        speed_calculation=SpeedCalculation.BORDER_SINGLE_SIDED,
+    )
+
+    data = pandas.merge(traj_data.data, velocity, on=[ID_COL, FRAME_COL])
+    data["velocity"] = shapely.points(data.v_x, data.v_y)
+
+    matrix = pandas.merge(
+        data, data, how="outer", on=FRAME_COL, suffixes=("", "_neighbor")
+    )
+    matrix = matrix[matrix.id != matrix.id_neighbor]
+
+    distance = np.linalg.norm(
+        shapely.get_coordinates(matrix.point)
+        - shapely.get_coordinates(matrix.point_neighbor),
+        axis=1,
+    )
+
+    e_v = (
+        shapely.get_coordinates(matrix.point)
+        - shapely.get_coordinates(matrix.point_neighbor)
+    ) / distance[:, np.newaxis]
+
+    v_rel = shapely.get_coordinates(matrix.velocity) - shapely.get_coordinates(
+        matrix.velocity_neighbor
+    ) / np.linalg.norm(
+        shapely.get_coordinates(matrix.velocity)
+        - shapely.get_coordinates(matrix.velocity_neighbor)
+    )
+
+    v_rel_norm = np.linalg.norm(v_rel, axis=1)
+
+    dot_product = np.sum(
+        np.array(np.array(e_v.tolist())) * np.array(np.array(v_rel.tolist())),
+        axis=1,
+    )
+
+    cos_alpha = dot_product / (distance * v_rel_norm)
+
+    ttc = np.full(matrix.shape[0], np.inf)
+
+    capital_a = (cos_alpha**2 - 1) * distance**2 + radius**2
+
+    # (0.5 * l_a + 0.5 * l_b)**2 in paper
+    sqrt_a_safe = np.where(
+        capital_a >= 0, np.sqrt(np.where(capital_a >= 0, capital_a, 0)), np.nan
+    )
+
+    valid_conditions = (
+        (capital_a >= 0)
+        & (-cos_alpha * distance - sqrt_a_safe >= 0)
+        & (v_rel_norm != 0)
+    )
+
+    ttc[valid_conditions] = (
+        -cos_alpha[valid_conditions] * distance[valid_conditions]
+        - np.sqrt(capital_a[valid_conditions])
+    ) / v_rel_norm[valid_conditions]
+
+    matrix[AVOIDANCE_COL] = tau_0 / ttc
+
+    avoidance = matrix.groupby(by=[ID_COL, FRAME_COL], as_index=False).agg(
+        avoidance=(AVOIDANCE_COL, "max")
+    )
+    return avoidance