system/ui: render model output with new ModelRenderer class (#35356)

render model output with new ModelRenderer class
pull/35363/head
Dean Lee 3 months ago committed by GitHub
parent 7511983ccb
commit 28da563386
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
  1. 15
      system/ui/onroad/augmented_road_view.py
  2. 360
      system/ui/onroad/model_renderer.py

@ -3,6 +3,7 @@ import pyray as rl
from cereal import messaging, log
from msgq.visionipc import VisionStreamType
from openpilot.system.ui.onroad.model_renderer import ModelRenderer
from openpilot.system.ui.widgets.cameraview import CameraView
from openpilot.system.ui.lib.application import gui_app
from openpilot.common.transformations.camera import DEVICE_CAMERAS, DeviceCameraConfig, view_frame_from_device_frame
@ -29,6 +30,8 @@ class AugmentedRoadView(CameraView):
self._last_rect_dims = (0.0, 0.0)
self._cached_matrix: np.ndarray | None = None
self.model_renderer = ModelRenderer()
def render(self, rect):
# Update calibration before rendering
self._update_calibration()
@ -41,6 +44,7 @@ class AugmentedRoadView(CameraView):
# - Path prediction
# - Lead vehicle indicators
# - Additional features
self.model_renderer.draw(rect, self.sm)
def _update_calibration(self):
# Update device camera if not already set
@ -112,12 +116,21 @@ class AugmentedRoadView(CameraView):
[0, 0, 1.0]
])
video_transform = np.array([
[zoom, 0.0, (w / 2 - x_offset) - (cx * zoom)],
[0.0, zoom, (h / 2 - y_offset) - (cy * zoom)],
[0.0, 0.0, 1.0]
])
self.model_renderer.set_transform(video_transform @ calib_transform)
return self._cached_matrix
if __name__ == "__main__":
gui_app.init_window("OnRoad Camera View")
sm = messaging.SubMaster(['deviceState', 'liveCalibration', 'roadCameraState'])
sm = messaging.SubMaster(["modelV2", "controlsState", "liveCalibration", "radarState", "deviceState",
"pandaStates", "carParams", "driverMonitoringState", "carState", "driverStateV2",
"roadCameraState", "wideRoadCameraState", "managerState", "selfdriveState", "longitudinalPlan"])
road_camera_view = AugmentedRoadView(sm, VisionStreamType.VISION_STREAM_ROAD)
try:
for _ in gui_app.render():

@ -0,0 +1,360 @@
import colorsys
import bisect
import numpy as np
import pyray as rl
from cereal import messaging, car
from openpilot.common.params import Params
CLIP_MARGIN = 500
MIN_DRAW_DISTANCE = 10.0
MAX_DRAW_DISTANCE = 100.0
THROTTLE_COLORS = [
rl.Color(0, 231, 130, 102), # Green with alpha 0.4
rl.Color(112, 247, 35, 89), # Lime with alpha 0.35
rl.Color(112, 247, 35, 0), # Transparent lime
]
NO_THROTTLE_COLORS = [
rl.Color(242, 242, 242, 102), # Light gray with alpha 0.4
rl.Color(242, 242, 242, 89), # Light gray with alpha 0.35
rl.Color(242, 242, 242, 0), # Transparent light gray
]
class ModelRenderer:
def __init__(self):
self._longitudinal_control = False
self._experimental_mode = False
self._blend_factor = 1.0
self._prev_allow_throttle = True
self._lane_line_probs = [0.0] * 4
self._road_edge_stds = [0.0] * 2
self._path_offset_z = 1.22
# Initialize empty polygon vertices
self._track_vertices = []
self._lane_line_vertices = [[] for _ in range(4)]
self._road_edge_vertices = [[] for _ in range(2)]
self._lead_vertices = [None, None]
# Transform matrix (3x3 for car space to screen space)
self._car_space_transform = np.zeros((3, 3))
self._transform_dirty = True
self._clip_region = None
# Get longitudinal control setting from car parameters
car_params = Params().get("CarParams")
if car_params:
cp = messaging.log_from_bytes(car_params, car.CarParams)
self._longitudinal_control = cp.openpilotLongitudinalControl
def set_transform(self, transform: np.ndarray):
self._car_space_transform = transform
self._transform_dirty = True
def draw(self, rect: rl.Rectangle, sm: messaging.SubMaster):
# Check if data is up-to-date
if not sm.valid['modelV2'] or not sm.valid['liveCalibration']:
return
# Set up clipping region
self._clip_region = rl.Rectangle(
rect.x - CLIP_MARGIN, rect.y - CLIP_MARGIN, rect.width + 2 * CLIP_MARGIN, rect.height + 2 * CLIP_MARGIN
)
# Update flags based on car state
self._experimental_mode = sm['selfdriveState'].experimentalMode
self._path_offset_z = sm['liveCalibration'].height[0]
if sm.updated['carParams']:
self._longitudinal_control = sm['carParams'].openpilotLongitudinalControl
# Get model and radar data
model = sm['modelV2']
radar_state = sm['radarState'] if sm.valid['radarState'] else None
lead_one = radar_state.leadOne if radar_state else None
render_lead_indicator = self._longitudinal_control and radar_state is not None
# Update model data when needed
if self._transform_dirty or sm.updated['modelV2'] or sm.updated['radarState']:
self._update_model(model, lead_one)
if render_lead_indicator:
self._update_leads(radar_state, model.position)
self._transform_dirty = False
# Draw elements
self._draw_lane_lines()
self._draw_path(sm, model, rect.height)
# Draw lead vehicles if available
if render_lead_indicator and radar_state:
lead_two = radar_state.leadTwo
if lead_one and lead_one.status:
self._draw_lead(lead_one, self._lead_vertices[0], rect)
if lead_two and lead_two.status and lead_one and (abs(lead_one.dRel - lead_two.dRel) > 3.0):
self._draw_lead(lead_two, self._lead_vertices[1], rect)
def _update_leads(self, radar_state, line):
"""Update positions of lead vehicles"""
leads = [radar_state.leadOne, radar_state.leadTwo]
for i, lead_data in enumerate(leads):
if lead_data and lead_data.status:
d_rel = lead_data.dRel
y_rel = lead_data.yRel
idx = self._get_path_length_idx(line, d_rel)
z = line.z[idx]
self._lead_vertices[i] = self._map_to_screen(d_rel, -y_rel, z + self._path_offset_z)
def _update_model(self, model, lead):
"""Update model visualization data based on model message"""
model_position = model.position
# Determine max distance to render
max_distance = np.clip(model_position.x[-1], MIN_DRAW_DISTANCE, MAX_DRAW_DISTANCE)
# Update lane lines
lane_lines = model.laneLines
line_probs = model.laneLineProbs
max_idx = self._get_path_length_idx(lane_lines[0], max_distance)
for i in range(4):
self._lane_line_probs[i] = line_probs[i]
self._lane_line_vertices[i] = self._map_line_to_polygon(
lane_lines[i], 0.025 * self._lane_line_probs[i], 0, max_idx
)
# Update road edges
road_edges = model.roadEdges
edge_stds = model.roadEdgeStds
for i in range(2):
self._road_edge_stds[i] = edge_stds[i]
self._road_edge_vertices[i] = self._map_line_to_polygon(road_edges[i], 0.025, 0, max_idx)
# Update path
if lead and lead.status:
lead_d = lead.dRel * 2.0
max_distance = np.clip(lead_d - min(lead_d * 0.35, 10.0), 0.0, max_distance)
max_idx = self._get_path_length_idx(model_position, max_distance)
self._track_vertices = self._map_line_to_polygon(model_position, 0.9, self._path_offset_z, max_idx, False)
def _draw_lane_lines(self):
"""Draw lane lines and road edges"""
for i in range(4):
# Skip if no vertices
if not self._lane_line_vertices[i]:
continue
# Draw lane line
alpha = np.clip(self._lane_line_probs[i], 0.0, 0.7)
color = rl.Color(255, 255, 255, int(alpha * 255))
self._draw_polygon(self._lane_line_vertices[i], color)
for i in range(2):
# Skip if no vertices
if not self._road_edge_vertices[i]:
continue
# Draw road edge
alpha = np.clip(1.0 - self._road_edge_stds[i], 0.0, 1.0)
color = rl.Color(255, 0, 0, int(alpha * 255))
self._draw_polygon(self._road_edge_vertices[i], color)
def _draw_path(self, sm, model, height):
"""Draw the path polygon with gradient based on acceleration"""
if not self._track_vertices:
return
if self._experimental_mode:
# Draw with acceleration coloring
acceleration = model.acceleration.x
max_len = min(len(self._track_vertices) // 2, len(acceleration))
# Create gradient colors for path sections
for i in range(max_len):
track_idx = max_len - i - 1 # flip idx to start from bottom right
track_y = self._track_vertices[track_idx][1]
# Skip points out of frame
if track_y < 0 or track_y > height:
continue
# Calculate color based on acceleration
lin_grad_point = (height - track_y) / height
# speed up: 120, slow down: 0
path_hue = max(min(60 + acceleration[i] * 35, 120), 0)
path_hue = int(path_hue * 100 + 0.5) / 100
saturation = min(abs(acceleration[i] * 1.5), 1)
lightness = self._map_val(saturation, 0.0, 1.0, 0.95, 0.62)
alpha = self._map_val(lin_grad_point, 0.75 / 2.0, 0.75, 0.4, 0.0)
# Use HSL to RGB conversion
color = self._hsla_to_color(path_hue / 360.0, saturation, lightness, alpha)
# TODO: This is simplified - a full implementation would create a gradient fill
segment = self._track_vertices[track_idx : track_idx + 2] + self._track_vertices[-track_idx - 2 : -track_idx]
self._draw_polygon(segment, color)
# Skip a point, unless next is last
i += 1 if i + 2 < max_len else 0
else:
# Draw with throttle/no throttle gradient
allow_throttle = sm['longitudinalPlan'].allowThrottle or not self._longitudinal_control
# Start transition if throttle state changes
if allow_throttle != self._prev_allow_throttle:
self._prev_allow_throttle = allow_throttle
self._blend_factor = max(1.0 - self._blend_factor, 0.0)
# Update blend factor
if self._blend_factor < 1.0:
self._blend_factor = min(self._blend_factor + 0.1, 1.0)
begin_colors = NO_THROTTLE_COLORS if allow_throttle else THROTTLE_COLORS
end_colors = THROTTLE_COLORS if allow_throttle else NO_THROTTLE_COLORS
# Blend colors based on transition
colors = [
self._blend_colors(begin_colors[0], end_colors[0], self._blend_factor),
self._blend_colors(begin_colors[1], end_colors[1], self._blend_factor),
self._blend_colors(begin_colors[2], end_colors[2], self._blend_factor),
]
self._draw_polygon(self._track_vertices, colors[0])
def _draw_lead(self, lead_data, vd, rect):
"""Draw lead vehicle indicator"""
if not vd:
return
speed_buff = 10.0
lead_buff = 40.0
d_rel = lead_data.dRel
v_rel = lead_data.vRel
# Calculate fill alpha
fill_alpha = 0
if d_rel < lead_buff:
fill_alpha = 255 * (1.0 - (d_rel / lead_buff))
if v_rel < 0:
fill_alpha += 255 * (-1 * (v_rel / speed_buff))
fill_alpha = min(fill_alpha, 255)
# Calculate size and position
sz = np.clip((25 * 30) / (d_rel / 3 + 30), 15.0, 30.0) * 2.35
x = np.clip(vd[0], 0.0, rect.width - sz / 2)
y = min(vd[1], rect.height - sz * 0.6)
g_xo = sz / 5
g_yo = sz / 10
# Draw glow
glow = [(x + (sz * 1.35) + g_xo, y + sz + g_yo), (x, y - g_yo), (x - (sz * 1.35) - g_xo, y + sz + g_yo)]
rl.draw_triangle_fan(glow, len(glow), rl.Color(218, 202, 37, 255))
# Draw chevron
chevron = [(x + (sz * 1.25), y + sz), (x, y), (x - (sz * 1.25), y + sz)]
rl.draw_triangle_fan(chevron, len(chevron), rl.Color(201, 34, 49, int(fill_alpha)))
@staticmethod
def _get_path_length_idx(line, path_height):
"""Get the index corresponding to the given path height"""
return bisect.bisect_right(line.x, path_height) - 1
def _map_to_screen(self, in_x, in_y, in_z):
"""Project a point in car space to screen space"""
input_pt = np.array([in_x, in_y, in_z])
pt = self._car_space_transform @ input_pt
if abs(pt[2]) < 1e-6:
return None
x = pt[0] / pt[2]
y = pt[1] / pt[2]
clip = self._clip_region
if x < clip.x or x > clip.x + clip.width or y < clip.y or y > clip.y + clip.height:
return None
return (x, y)
def _map_line_to_polygon(self, line, y_off, z_off, max_idx, allow_invert=True):
"""Convert a 3D line to a 2D polygon for drawing"""
line_x = line.x
line_y = line.y
line_z = line.z
left_points = []
right_points = []
for i in range(max_idx + 1):
# Skip points with negative x (behind camera)
if line_x[i] < 0:
continue
left = self._map_to_screen(line_x[i], line_y[i] - y_off, line_z[i] + z_off)
right = self._map_to_screen(line_x[i], line_y[i] + y_off, line_z[i] + z_off)
if left and right:
# Check for inversion when going over hills
if not allow_invert and left_points and left[1] > left_points[-1][1]:
continue
left_points.append(left)
right_points.append(right)
if not left_points:
return []
return left_points + right_points[::-1]
def _draw_polygon(self, points, color):
# TODO: Enhance polygon drawing to support even-odd fill rule efficiently, as Raylib lacks native support.
# Use a faster triangulation algorithm (e.g., ear clipping) or GPU shader for
# efficient rendering of lane lines, road edges, and path polygons.
if len(points) <= 8:
rl.draw_triangle_fan(points, len(points), color)
else:
for i in range(1, len(points) - 1):
rl.draw_triangle(points[0], points[i], points[i + 1], color)
for i in range(len(points)):
rl.draw_line_ex(points[i], points[(i + 1) % len(points)], 1.5, color)
@staticmethod
def _map_val(x, x0, x1, y0, y1):
"""Map value x from range [x0, x1] to range [y0, y1]"""
return y0 + (y1 - y0) * ((x - x0) / (x1 - x0)) if x1 != x0 else y0
@staticmethod
def _hsla_to_color(h, s, l, a):
"""Convert HSLA color to Raylib Color using colorsys module"""
# colorsys uses HLS format (Hue, Lightness, Saturation)
r, g, b = colorsys.hls_to_rgb(h, l, s)
# Ensure values are in valid range
r_val = max(0, min(255, int(r * 255)))
g_val = max(0, min(255, int(g * 255)))
b_val = max(0, min(255, int(b * 255)))
a_val = max(0, min(255, int(a * 255)))
return rl.Color(r_val, g_val, b_val, a_val)
@staticmethod
def _blend_colors(start, end, t):
"""Blend between two colors with factor t"""
if t >= 1.0:
return end
return rl.Color(
int((1 - t) * start.r + t * end.r),
int((1 - t) * start.g + t * end.g),
int((1 - t) * start.b + t * end.b),
int((1 - t) * start.a + t * end.a),
)
Loading…
Cancel
Save