openpilot is an open source driver assistance system. openpilot performs the functions of Automated Lane Centering and Adaptive Cruise Control for over 200 supported car makes and models.
You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

242 lines
9.6 KiB

import numpy as np
import pyray as rl
from dataclasses import dataclass
from openpilot.selfdrive.ui.ui_state import ui_state, UI_BORDER_SIZE
from openpilot.system.ui.lib.application import gui_app, Widget
# Default 3D coordinates for face keypoints as a NumPy array
DEFAULT_FACE_KPTS_3D = np.array([
[-5.98, -51.20, 8.00], [-17.64, -49.14, 8.00], [-23.81, -46.40, 8.00], [-29.98, -40.91, 8.00],
[-32.04, -37.49, 8.00], [-34.10, -32.00, 8.00], [-36.16, -21.03, 8.00], [-36.16, 6.40, 8.00],
[-35.47, 10.51, 8.00], [-32.73, 19.43, 8.00], [-29.30, 26.29, 8.00], [-24.50, 33.83, 8.00],
[-19.01, 41.37, 8.00], [-14.21, 46.17, 8.00], [-12.16, 47.54, 8.00], [-4.61, 49.60, 8.00],
[4.99, 49.60, 8.00], [12.53, 47.54, 8.00], [14.59, 46.17, 8.00], [19.39, 41.37, 8.00],
[24.87, 33.83, 8.00], [29.67, 26.29, 8.00], [33.10, 19.43, 8.00], [35.84, 10.51, 8.00],
[36.53, 6.40, 8.00], [36.53, -21.03, 8.00], [34.47, -32.00, 8.00], [32.42, -37.49, 8.00],
[30.36, -40.91, 8.00], [24.19, -46.40, 8.00], [18.02, -49.14, 8.00], [6.36, -51.20, 8.00],
[-5.98, -51.20, 8.00],
], dtype=np.float32)
# UI constants
BTN_SIZE = 192
IMG_SIZE = 144
ARC_LENGTH = 133
ARC_THICKNESS_DEFAULT = 6.7
ARC_THICKNESS_EXTEND = 12.0
SCALES_POS = np.array([0.9, 0.4, 0.4], dtype=np.float32)
SCALES_NEG = np.array([0.7, 0.4, 0.4], dtype=np.float32)
ARC_POINT_COUNT = 37 # Number of points in the arc
ARC_ANGLES = np.linspace(0.0, np.pi, ARC_POINT_COUNT, dtype=np.float32)
@dataclass
class ArcData:
"""Data structure for arc rendering parameters."""
x: float
y: float
width: float
height: float
thickness: float
class DriverStateRenderer(Widget):
def __init__(self):
super().__init__()
# Initial state with NumPy arrays
self.face_kpts_draw = DEFAULT_FACE_KPTS_3D.copy()
self.is_active = False
self.is_rhd = False
self.dm_fade_state = 0.0
self.state_updated = False
self.last_rect: rl.Rectangle = rl.Rectangle(0, 0, 0, 0)
self.driver_pose_vals = np.zeros(3, dtype=np.float32)
self.driver_pose_diff = np.zeros(3, dtype=np.float32)
self.driver_pose_sins = np.zeros(3, dtype=np.float32)
self.driver_pose_coss = np.zeros(3, dtype=np.float32)
self.face_keypoints_transformed = np.zeros((DEFAULT_FACE_KPTS_3D.shape[0], 2), dtype=np.float32)
self.position_x: float = 0.0
self.position_y: float = 0.0
self.h_arc_data = None
self.v_arc_data = None
# Pre-allocate drawing arrays
self.face_lines = [rl.Vector2(0, 0) for _ in range(len(DEFAULT_FACE_KPTS_3D))]
self.h_arc_lines = [rl.Vector2(0, 0) for _ in range(ARC_POINT_COUNT)]
self.v_arc_lines = [rl.Vector2(0, 0) for _ in range(ARC_POINT_COUNT)]
# Load the driver face icon
self.dm_img = gui_app.texture("icons/driver_face.png", IMG_SIZE, IMG_SIZE)
# Colors
self.white_color = rl.Color(255, 255, 255, 255)
self.arc_color = rl.Color(26, 242, 66, 255)
self.engaged_color = rl.Color(26, 242, 66, 255)
self.disengaged_color = rl.Color(139, 139, 139, 255)
def _render(self, rect):
if not self._is_visible(ui_state.sm):
return
self._update_state(ui_state.sm, rect)
if not self.state_updated:
return
# Set opacity based on active state
opacity = 0.65 if self.is_active else 0.2
# Draw background circle
rl.draw_circle(int(self.position_x), int(self.position_y), BTN_SIZE // 2, rl.Color(0, 0, 0, 70))
# Draw face icon
icon_pos = rl.Vector2(self.position_x - self.dm_img.width // 2, self.position_y - self.dm_img.height // 2)
rl.draw_texture_v(self.dm_img, icon_pos, rl.Color(255, 255, 255, int(255 * opacity)))
# Draw face outline
self.white_color.a = int(255 * opacity)
rl.draw_spline_linear(self.face_lines, len(self.face_lines), 5.2, self.white_color)
# Set arc color based on engaged state
self.arc_color = self.engaged_color if ui_state.engaged else self.disengaged_color
self.arc_color.a = int(0.4 * 255 * (1.0 - self.dm_fade_state)) # Fade out when inactive
# Draw arcs
if self.h_arc_data:
rl.draw_spline_linear(self.h_arc_lines, len(self.h_arc_lines), self.h_arc_data.thickness, self.arc_color)
if self.v_arc_data:
rl.draw_spline_linear(self.v_arc_lines, len(self.v_arc_lines), self.v_arc_data.thickness, self.arc_color)
def _is_visible(self, sm):
"""Check if the visualization should be rendered."""
return (sm.recv_frame['driverStateV2'] > ui_state.started_frame and
sm.seen['driverMonitoringState'] and
sm['selfdriveState'].alertSize == 0)
def _update_state(self, sm, rect):
"""Update the driver monitoring state based on model data"""
if not sm.updated["driverMonitoringState"]:
if self.state_updated and (rect.x != self.last_rect.x or rect.y != self.last_rect.y or
rect.width != self.last_rect.width or rect.height != self.last_rect.height):
self._pre_calculate_drawing_elements(rect)
return
# Get monitoring state
dm_state = sm["driverMonitoringState"]
self.is_active = dm_state.isActiveMode
self.is_rhd = dm_state.isRHD
# Update fade state (smoother transition between active/inactive)
fade_target = 0.0 if self.is_active else 0.5
self.dm_fade_state = np.clip(self.dm_fade_state + 0.2 * (fade_target - self.dm_fade_state), 0.0, 1.0)
# Get driver orientation data from appropriate camera
driverstate = sm["driverStateV2"]
driver_data = driverstate.rightDriverData if self.is_rhd else driverstate.leftDriverData
driver_orient = driver_data.faceOrientation
# Update pose values with scaling and smoothing
driver_orient = np.array(driver_orient)
scales = np.where(driver_orient < 0, SCALES_NEG, SCALES_POS)
v_this = driver_orient * scales
self.driver_pose_diff = np.abs(self.driver_pose_vals - v_this)
self.driver_pose_vals = 0.8 * v_this + 0.2 * self.driver_pose_vals # Smooth changes
# Apply fade to rotation and compute sin/cos
rotation_amount = self.driver_pose_vals * (1.0 - self.dm_fade_state)
self.driver_pose_sins = np.sin(rotation_amount)
self.driver_pose_coss = np.cos(rotation_amount)
# Create rotation matrix for 3D face model
sin_y, sin_x, sin_z = self.driver_pose_sins
cos_y, cos_x, cos_z = self.driver_pose_coss
r_xyz = np.array(
[
[cos_x * cos_z, cos_x * sin_z, -sin_x],
[-sin_y * sin_x * cos_z - cos_y * sin_z, -sin_y * sin_x * sin_z + cos_y * cos_z, -sin_y * cos_x],
[cos_y * sin_x * cos_z - sin_y * sin_z, cos_y * sin_x * sin_z + sin_y * cos_z, cos_y * cos_x],
]
)
# Transform face keypoints using vectorized matrix multiplication
self.face_kpts_draw = DEFAULT_FACE_KPTS_3D @ r_xyz.T
self.face_kpts_draw[:, 2] = self.face_kpts_draw[:, 2] * (1.0 - self.dm_fade_state) + 8 * self.dm_fade_state
# Pre-calculate the transformed keypoints
kp_depth = (self.face_kpts_draw[:, 2] - 8) / 120.0 + 1.0
self.face_keypoints_transformed = self.face_kpts_draw[:, :2] * kp_depth[:, None]
# Pre-calculate all drawing elements
self._pre_calculate_drawing_elements(rect)
self.state_updated = True
def _pre_calculate_drawing_elements(self, rect):
"""Pre-calculate all drawing elements based on the current rectangle"""
# Calculate icon position (bottom-left or bottom-right)
width, height = rect.width, rect.height
offset = UI_BORDER_SIZE + BTN_SIZE // 2
self.position_x = rect.x + (width - offset if self.is_rhd else offset)
self.position_y = rect.y + height - offset
# Pre-calculate the face lines positions
positioned_keypoints = self.face_keypoints_transformed + np.array([self.position_x, self.position_y])
for i in range(len(positioned_keypoints)):
self.face_lines[i].x = positioned_keypoints[i][0]
self.face_lines[i].y = positioned_keypoints[i][1]
# Calculate arc dimensions based on head rotation
delta_x = -self.driver_pose_sins[1] * ARC_LENGTH / 2.0 # Horizontal movement
delta_y = -self.driver_pose_sins[0] * ARC_LENGTH / 2.0 # Vertical movement
# Horizontal arc
h_width = abs(delta_x)
self.h_arc_data = self._calculate_arc_data(
delta_x, h_width, self.position_x, self.position_y - ARC_LENGTH / 2,
self.driver_pose_sins[1], self.driver_pose_diff[1], is_horizontal=True
)
# Vertical arc
v_height = abs(delta_y)
self.v_arc_data = self._calculate_arc_data(
delta_y, v_height, self.position_x - ARC_LENGTH / 2, self.position_y,
self.driver_pose_sins[0], self.driver_pose_diff[0], is_horizontal=False
)
def _calculate_arc_data(
self, delta: float, size: float, x: float, y: float, sin_val: float, diff_val: float, is_horizontal: bool
):
"""Calculate arc data and pre-compute arc points."""
if size <= 0:
return None
thickness = ARC_THICKNESS_DEFAULT + ARC_THICKNESS_EXTEND * min(1.0, diff_val * 5.0)
start_angle = (90 if sin_val > 0 else -90) if is_horizontal else (0 if sin_val > 0 else 180)
x = min(x + delta, x) if is_horizontal else x
y = y if is_horizontal else min(y + delta, y)
arc_data = ArcData(
x=x,
y=y,
width=size if is_horizontal else ARC_LENGTH,
height=ARC_LENGTH if is_horizontal else size,
thickness=thickness,
)
# Pre-calculate arc points
angles = ARC_ANGLES + np.deg2rad(start_angle)
center_x = x + arc_data.width / 2
center_y = y + arc_data.height / 2
radius_x = arc_data.width / 2
radius_y = arc_data.height / 2
x_coords = center_x + np.cos(angles) * radius_x
y_coords = center_y + np.sin(angles) * radius_y
arc_lines = self.h_arc_lines if is_horizontal else self.v_arc_lines
for i, (x_coord, y_coord) in enumerate(zip(x_coords, y_coords, strict=True)):
arc_lines[i].x = x_coord
arc_lines[i].y = y_coord
return arc_data