class Traffic:
def __init__(self, roadNetwork, vehicles, size, totalTime):
self.roadNetwork = roadNetwork
self.vehicles = vehicles
self.size = size
self.supportVehicleSize = 1
self.exportVehicleIndices = [-1]*30
self.vehicleSequences = []
self.exportfp = open("trafficExport.data","w")
self.totalTime = totalTime
self.vehicleTypes = VehicleTypes()
self.grid = np.zeros(size)
self.drawNetwork()
self.drawVehicles()
def addVehicle(self, lanes, vehicleTypes):
if vehicleTypes:
vehicleType, _ = self.vehicleTypes.sample()
else:
vehicleType = 'CAR'
if not lanes:
vehicle = Vehicle(vehicleType, self.roadNetwork, 0, SimpleLaneMovement())
else:
l = [1,2]
laneNo = l[int(random.random()*2)]
vehicle = Vehicle(vehicleType, self.roadNetwork, 0, SimpleLaneMovement(), 4, laneNo)
return vehicle
def addVehicleWithSpeed(self):
speed = [1,2,3,4,5]
speed_prob = [0.1,0.3,0.3,0.2,0.1]
lane = [0,1,2,3]
lane_prob = [0.2,0.3,0.3,0.2]
speed = np.random.choice(speed, 1, p=speed_prob)[0]
lane = np.random.choice(lane, 1, p=lane_prob)[0]
vehicleType, _ = self.vehicleTypes.sample()
vehicle = Vehicle(vehicleType, self.roadNetwork, 0, SimpleLaneMovement(), speed, lane)
return vehicle
def drawNetwork(self):
edges = self.roadNetwork.edges
nodes = self.roadNetwork.nodes
for edge in edges:
nodeI = edge.node1
nodeJ = edge.node2
width = edge.width
### Prastutaniki angle em ledhu, only straight lines., commented because I wanted the road to cover the entire the screen, its ok to uncomment if you want.
#cv2.line(self.grid, (int(nodeI.x-width/2), nodeI.y), (int(nodeI.x-width/2), nodeJ.y), (255,0,0),1)
#cv2.line(self.grid, (int(nodeI.x+width/2), nodeI.y), (int(nodeI.x+width/2), nodeJ.y), (255,0,0),1)
## Drawing a center white line for SUMO purpose
cv2.line(self.grid, (nodeI.x, nodeI.y), (nodeJ.x, nodeJ.y), (255,255,255), 1)
### if Edges have lanes draw the lanes.
def drawNetworkWithHomography(self, H):
edges = self.roadNetwork.edges
nodes = self.roadNetwork.nodes
for edge in edges:
nodeI = edge.node1
nodeJ = edge.node2
width = edge.width
### Prastutaniki angle em ledhu, only straight lines.
pts1 = np.array([[int(nodeI.x-width/2), nodeI.y], [int(nodeI.x-width/2), nodeJ.y]])
pts2 = np.array([[int(nodeI.x+width/2), nodeI.y], [int(nodeI.x+width/2), nodeJ.y]])
pts1 = transformSetOfPointsAndReturn(pts1, H)
pts2 = transformSetOfPointsAndReturn(pts2, H)
cv2.line(self.grid, (pts1[0][0], pts1[0][1]), (pts1[1][0], pts1[1][1]), (255,0,0), 1)
cv2.line(self.grid, (pts2[0][0], pts2[0][1]), (pts2[1][0], pts2[1][1]), (255,0,0), 1)
def get5nearestNeighbours(self, vehicleIndex):
dist = {}
a = np.array([self.vehicles[vehicleIndex].curX, self.vehicles[vehicleIndex].curY])
for i,vehicle in enumerate(self.vehicles):
if (i == vehicleIndex) or (len(vehicle.track)) < 2 or (vehicle.track[-2][0] == -1 and vehicle.track[-2][1] == -1):
continue
b = np.array([vehicle.curX, vehicle.curY])
d = np.linalg.norm(a-b)
if len(dist.keys()) < 5:
dist[d] = i
continue
mx = max(dist.keys())
if mx > d:
dist.pop(mx,None)
return dist
def updateExportVehiclesIndices(self):
l = []
for i in self.exportVehicleIndices:
v1 = []
print i,
if i == -1 or len(self.vehicles[i].track) < 2 or (self.vehicles[i].track[-2][0] == -1 and self.vehicles[i].track[-2][1] == -1):
v1 += [-1]*17
else:
v1 += self.vehicleTypes.oneHotEncoding(self.vehicles[i].cl)
x = self.vehicles[i].track[-2][0]
y = self.vehicles[i].track[-2][1]
v1 += [x, y]
print v1
neighs = self.get5nearestNeighbours(i)
print neighs
for n in neighs.keys():
if self.vehicles[neighs[n]].track[-2][0] != -1 and self.vehicles[neighs[n]].track[-2][1] != -1:
v1 += [x-self.vehicles[neighs[n]].track[-2][0], y-self.vehicles[neighs[n]].track[-2][1]]
else:
v1 +=[-1,-1]
for k in range(len(neighs.keys()),5):
v1 += [-1,-1]
if self.vehicles[i].track[-2][0] == -1 and self.vehicles[i].track[-2][1] == -1:
v1 += [0,0]
else:
v1 += [self.vehicles[i].track[-2][0] - self.vehicles[i].curX, self.vehicles[i].track[-2][1] - self.vehicles[i].curY]
print v1,
l.append(v1)
print "\n",
for i, vehicle in enumerate(self.vehicles):
#print "Guru ", i, vehicle.numberOfEdgesCompleted
if vehicle.numberOfEdgesCompleted == 1 and i in self.exportVehicleIndices:
self.exportVehicleIndices[self.exportVehicleIndices.index(i)] = -1
elif i not in self.exportVehicleIndices:
if -1 in self.exportVehicleIndices:
self.exportVehicleIndices[self.exportVehicleIndices.index(-1)] = i
def drawVehicles(self):
for vehicle in self.vehicles:
if vehicle.curX == -1 and vehicle.curY == -1:
continue
vehi_size = self.vehicleTypes.getSize(vehicle.cl)
if self.supportVehicleSize:
#print "Rectangles to be printed are : ", vehicle.curX-vehi_size[0]/2,vehicle.curY-vehi_size[1]/2, " " , vehicle.curX+vehi_size[0]/2, vehicle.curY+vehi_size[1]/2
cv2.rectangle(self.grid,(vehicle.curX-vehi_size[0]/2,vehicle.curY-vehi_size[1]/2), (vehicle.curX+vehi_size[0]/2,vehicle.curY+vehi_size[1]/2), vehicle.color, thickness=cv2.cv.CV_FILLED)
cv2.putText(self.grid, str(vehicle.identity), (vehicle.curX-vehi_size[0]/2,vehicle.curY-vehi_size[1]/2), cv2.FONT_HERSHEY_SIMPLEX, 0.75, (255,0,0), 1, cv2.CV_AA)
else:
self.grid[vehicle.curY, vehicle.curX] = [255,0,0]
def drawVehiclesWithHomography(self, H):
for vehicle in self.vehicles:
if vehicle.curX == -1 and vehicle.curY == -1:
continue
vehi_size = self.vehicleTypes.getSize(vehicle.cl)
if self.supportVehicleSize:
pts = np.array([[vehicle.curX-vehi_size[0]/2,vehicle.curY-vehi_size[1]/2],[vehicle.curX+vehi_size[0]/2, vehicle.curY-vehi_size[1]/2],[vehicle.curX+vehi_size[0]/2, vehicle.curY+vehi_size[1]/2], [vehicle.curX-vehi_size[0]/2, vehicle.curY+vehi_size[1]/2]], dtype=float)
pts = transformPoly(pts, H)
self.grid = drawPoly(self.grid, pts, vehicle.color)
#cv2.rectangle(self.grid,(vehicle.curX-vehi_size[0]/2,vehicle.curY-vehi_size[1]/2), (vehicle.curX+vehi_size[0]/2,vehicle.curY+vehi_size[1]/2), (0,255,0), 1)
else:
self.grid[vehicle.curY, vehicle.curX] = [255,0,0]
def updateGrid(self):
self.grid = np.zeros(self.size)
self.drawNetwork()
self.drawVehicles()
def updateGridWithHomography(self, H, perspective_road_points, straight_road_points):
self.grid = np.zeros(self.size)
#self.grid = drawPoly(self.grid, perspective_road_points)
#self.grid = drawPoly(self.grid, straight_road_points)
self.drawNetworkWithHomography(H)
self.drawVehiclesWithHomography(H)
def simulateAndVisualize(self):
t = 0
while(t < self.totalTime):
print "=========== Frame no : ", t, " =========="
for vehicle in self.vehicles:
vehicle.move(t, self.grid, self.vehicles)
self.updateGrid()
t += 1
cv2.imshow("sim", self.grid)
if cv2.waitKey(33) == 27:
break
#if random.random() < 0.1:
# self.vehicles.append(self.addVehicle())
# cv2.waitKey(0)
cv2.destroyAllWindows()
def simulateAndVisualizeWithHomography(self, H, perspective_road_points, straight_road_points):
t = 0
video ="/Users/gramaguru/Desktop/car_videos/sing.mp4"
cap = cv2.VideoCapture(video);
while(t < self.totalTime):
ret, frame = cap.read()
print frame.shape
frame = misc.imresize(frame, (384,512,3))
for vehicle in self.vehicles:
vehicle.move(t, self.grid, self.vehicles)
self.updateGridWithHomography(H, perspective_road_points, straight_road_points)
t += 1
cv2.imshow("orginal", frame)
cv2.imshow("sim", self.grid)
if cv2.waitKey(33) == 27:
break
#if random.random() < 0.1:
# self.vehicles.append(self.addVehicle())
# cv2.waitKey(0)
cv2.destroyAllWindows()
def simulateContinuousTraffic(self, total_time, speed=False, lanes=False, vehicleTypes = False):
t = 0
while(t < total_time):
for vehicle in self.vehicles:
vehicle.move(t, self.grid, self.vehicles)
self.updateGrid()
#cv2.imshow("sim", self.grid)
#if cv2.waitKey(33) == 27:
# break
#if t % 5 == 0:
if random.random() <= 0.2:
if speed:
self.vehicles.append(self.addVehicleWithSpeed())
else:
self.vehicles.append(self.addVehicle(lanes,vehicleTypes))
#self.vehicleSequences.append(self.vehicleTypes.oneHotEncoding(vehicle.cl) + self.roadNetwork.edges[vehicle.numberOfEdgesCompleted].oneHotEncoding(vehicle.laneNo))
if speed:
self.vehicleSequences.append(vehicle.cl + " " + str(vehicle.laneNo) + " " + str(vehicle.speed))
else:
self.vehicleSequences.append(vehicle.cl + " " + str(vehicle.laneNo))
#print "Guru ", self.vehicleTypes.oneHotEncoding(vehicle.cl) + self.roadNetwork.edges[vehicle.numberOfEdgesCompleted].oneHotEncoding(vehicle.laneNo)
else:
if speed:
self.vehicleSequences.append(("None 0 0"))
else:
self.vehicleSequences.append(("None 0"))
#self.updateExportVehiclesIndices()
t += 1
#print self.vehicleSequences
self.exportGeneratedVehicles(speed)
cv2.destroyAllWindows()
def generateTrafficUsingNN(self, total_time, start_seq):
# Parse command line arguments
t = 0
#decoder = torch.load("/Users/gramaguru/ComputerVision/computer-vision/Traffic/IndianTraffic/trafficGenerator/trafficNextPrediction/exportGeneratedAlphabets.pt")
#decoder = torch.load("/Users/gramaguru/ComputerVision/computer-vision/Traffic/IndianTraffic/trafficGenerator/trafficNextPrediction/traffic_seq_road_only_cars.pt")
decoder = torch.load("/Users/gramaguru/ComputerVision/computer-vision/Traffic/IndianTraffic/trafficGenerator/trafficNextPrediction/traffic_seq_roadNetwork_speed.export_51.pt")
parameter_len = 3
seq = generate_couple(decoder, start_seq, total_time, 0.8, False, parameter_len)
print seq
while(t < total_time):
v = seq[parameter_len*t:parameter_len*t+parameter_len]
laneNo = int(v[1])
if parameter_len == 2:
speed = 4
elif parameter_len == 3:
speed = int(v[2])
if v[0] != 'n' and speed != 0:
self.vehicles.append(Vehicle(self.vehicleTypes.map[v[0]], self.roadNetwork, 0, SimpleLaneMovement(), speed, laneNo))
for vehicle in self.vehicles:
vehicle.move(t, self.grid, self.vehicles)
self.updateGrid()
cv2.imshow("NNsim", self.grid)
if cv2.waitKey(33) == 27:
break
t += 1
def simulateAndExport(self):
t = 0
while(t < self.totalTime):
for vehicle in self.vehicles:
vehicle.move(t, self.grid, self.vehicles)
self.updateGrid()
t += 1
self.export()
def exportGeneratedVehicles(self, speed= False):
#exportGeneratedSeq = raw_input()
if speed:
exportGeneratedSeq = "traffic_seq_prob_dist_speed.txt";
else:
exportGeneratedSeq = "traffic_seq_prob_dist.txt";
print self.vehicleSequences
fp = open(exportGeneratedSeq, "w")
fp.write(str(self.vehicleSequences))
fp.close()
def export(self):
exportFileName = "traffic_" + str(len(self.roadNetwork.edges)) + "_" + str(len(self.vehicles)) + "_" + str(time.time()) + ".txt"
fp = open(exportFileName, "w")
trafficSummary = str(self.grid.shape[0]) + " " + str(self.grid.shape[1]) + " " + str(self.totalTime) + "\n"
fp.write(trafficSummary)
fp.write(str(len(self.roadNetwork.edges)) + "\n")
for edge in self.roadNetwork.edges:
fp.write(str(edge.node1.x) + " " + str(edge.node1.y) + " " + str(edge.node2.x) + " " + str(edge.node2.y) + " " + str(edge.width) + "\n")
fp.write(str(len(self.vehicles)) + "\n")
for vehicle in self.vehicles:
fp.write(vehicle.cl + "\n")
tracklen = len(vehicle.track)
diff = 0
if tracklen < self.totalTime:
diff = self.totalTime - tracklen
for t in range(self.totalTime):
if t < diff:
fp.write(str(t) + " " + str(-1) + " " + str(-1) + "\n")
continue
fp.write(str(t) + " " + str(vehicle.track[t-diff][0]) + " " + str(vehicle.track[t-diff][1]) + "\n")
fp.close()
def getSpeeds(self, vehicles):
speeds = {}
for vehicle in vehicles:
if len(vehicle.track) <= 1 or vehicle.track[-1][0] == -1 or vehicle.track[-1][1] == -1 or vehicle.track[-2][0] == -1 or vehicle.track[-2][1] == -1:
speeds[vehicle.identity] = [0.0,vehicle.speed]
print "Guru Guru", vehicle.identity, vehicle.speed
else:
speeds[vehicle.identity] = [vehicle.track[-2][0] - vehicle.track[-1][0], vehicle.track[-2][1] - vehicle.track[-1][1]]
return speeds
def get_vehicles_lanes(self, vehicles):
lanes_frames = {}
for vehicle in vehicles:
if vehicle.curX not in lanes_frames.keys():
lanes_frames[vehicle.curX] = [[vehicle.identity, vehicle.curX, vehicle.curY]]
else:
lanes_frames[vehicle.curX].append([vehicle.identity, vehicle.curX, vehicle.curY])
return lanes_frames
def get_vehicles_lanes_dist(self, vehicles, lanes):
lanes_frames = {}
for vehicle in vehicles:
lane = min(lanes, key=lambda x:abs(x-vehicle.curX))
if lane not in lanes_frames.keys():
lanes_frames[lane] = [[vehicle.identity, vehicle.curX, vehicle.curY]]
else:
lanes_frames[lane].append([vehicle.identity, vehicle.curX, vehicle.curY])
return lanes_frames
def getVehiclesInScreen(self, vehicles):
filtered_vehicles = []
for vehicle in vehicles:
if vehicle.curX == -1 or vehicle.curY == -1:
continue
filtered_vehicles.append(vehicle)
return filtered_vehicles
class Traffic:
def __init__(self, roadNetwork, vehicles, size, totalTime):
self.roadNetwork = roadNetwork
self.vehicles = vehicles
self.size = size
self.supportVehicleSize = 1
self.exportVehicleIndices = [-1] * 30
self.vehicleSequences = []
self.exportfp = open("trafficExport.data", "w")
self.totalTime = totalTime
self.vehicleTypes = VehicleTypes()
self.grid = np.zeros(size)
self.drawNetwork()
self.drawVehicles()
self.vehicleTypes = VehicleTypes()
self.dataLoader = SumoDataLoader(
"", 0.1, 0.1, 1, 5, 'log', 'infer',
True) # Data loader for getting the feature vector.
'''
self.max_xval = np.array([8.01000000e+01, 1.66500000e+01, 4.00000000e+00, 6.42890000e+02, 8.01000000e+01, 1.80500000e+01, -1.00000000e+00, 8.01000000e+01 ,1.80400000e+01, 6.34310000e+02, 8.01000000e+01,1.78700000e+01, -1.00000000e-02, 8.01000000e+01, 1.80400000e+01, 6.12900000e+02, 8.01000000e+01, 1.79600000e+01, -2.00000000e-02, 8.01000000e+01, 1.80400000e+01])
self.max_yval = 15.99
self.min_xval = np.array([-80.1, 0., 0., -1., -80.1, -1., -641.88, -80.1, -1., -1., -80.1, -1., -592.8, -80.1, -1., -1., -80.1, -1., -639.8, -80.1, -1.])
self.min_yval = 10
'''
self.min_xval = np.array([
-80.1, 0., 0., -1., -80.1, -1., -869.72, -80.1, -1., -1., -80.1,
-1., -858.51, -80.1, -1., -1., -80.1, -1., -826.39, -80.1, -1.
])
self.max_xval = np.array([
8.01000000e+01, 4.01600000e+01, 4.00000000e+00, 8.76950000e+02,
8.01000000e+01, 3.73800000e+01, -1.00000000e+00, 8.01000000e+01,
4.03500000e+01, 8.32860000e+02, 8.01000000e+01, 3.87300000e+01,
-1.00000000e-02, 8.01000000e+01, 4.01600000e+01, 8.25250000e+02,
8.01000000e+01, 3.89900000e+01, -1.00000000e-02, 8.01000000e+01,
3.80100000e+01
])
self.min_yval = 1.0
self.max_yval = 40.35
json_file = open('kerasmodels/model_speeds_4.json', 'r')
loaded_model_json = json_file.read()
json_file.close()
self.loaded_model = model_from_json(loaded_model_json)
# load weights into new model
self.loaded_model.load_weights("kerasmodels/model_speeds_4.h5")
self.dataLoader1 = nn_data(
"", 0.1, 0.1, 1, 5, 'log', 'infer',
True) # Data loader for getting the feature vector.
self.min_xval1 = np.array([
-5.641, -23.108, -116.837, -942.203, -9.548, -23.108, -124.331,
-1435.677, -7.502, -23.108, -135.313, -1485.024, -9.548, -21.737,
-144.973, -1553.565, -9.548, -22.14
])
self.max_yval1 = np.array([10.017, 13.639])
self.max_xval1 = np.array([
8.558, 12.448, 124.82, 422.733, 8.304, 13.639, 135.174, 517.,
10.017, 12.448, 140.7, 1531.639, 6.777, 13.639, 154.802, 1553.565,
8.558, 12.448
])
self.min_yval1 = np.array([-9.548, -26.801])
json_file = open('kerasmodels/model_us101.json', 'r')
loaded_model_json = json_file.read()
json_file.close()
self.loaded_model1 = model_from_json(loaded_model_json)
# load weights into new model
self.loaded_model1.load_weights("kerasmodels/model_us101.h5")
def addVehicle(self, lanes, vehicleTypes):
if vehicleTypes:
vehicleType, _ = self.vehicleTypes.sample()
else:
vehicleType = 'CAR'
if not lanes:
vehicle = Vehicle(vehicleType, self.roadNetwork, 0,
SimpleLaneMovement())
else:
l = [1, 2]
laneNo = l[int(random.random() * 2)]
vehicle = Vehicle(vehicleType, self.roadNetwork, 0,
SimpleLaneMovement(), 4, laneNo)
return vehicle
def addVehicleWithSpeed(self):
speed = [1, 2, 3, 4, 5]
speed_prob = [0.1, 0.3, 0.3, 0.2, 0.1]
lane = [0, 1, 2, 3]
lane_prob = [0.2, 0.3, 0.3, 0.2]
speed = np.random.choice(speed, 1, p=speed_prob)[0]
lane = np.random.choice(lane, 1, p=lane_prob)[0]
vehicleType, _ = self.vehicleTypes.sample()
vehicle = Vehicle(vehicleType, self.roadNetwork, 0,
SimpleLaneMovement(), speed, lane)
return vehicle
def drawNetwork(self):
edges = self.roadNetwork.edges
nodes = self.roadNetwork.nodes
for edge in edges:
nodeI = edge.node1
nodeJ = edge.node2
width = edge.width
### Prastutaniki angle em ledhu, only straight lines., commented because I wanted the road to cover the entire the screen, its ok to uncomment if you want.
#cv2.line(self.grid, (int(nodeI.x-width/2), nodeI.y), (int(nodeI.x-width/2), nodeJ.y), (255,0,0),1)
#cv2.line(self.grid, (int(nodeI.x+width/2), nodeI.y), (int(nodeI.x+width/2), nodeJ.y), (255,0,0),1)
## Drawing a center white line for SUMO purpose
cv2.line(self.grid, (nodeI.x, nodeI.y), (nodeJ.x, nodeJ.y),
(255, 255, 255), 1)
### if Edges have lanes draw the lanes.
def drawNetworkWithHomography(self, H):
edges = self.roadNetwork.edges
nodes = self.roadNetwork.nodes
for edge in edges:
nodeI = edge.node1
nodeJ = edge.node2
width = edge.width
### Prastutaniki angle em ledhu, only straight lines.
pts1 = np.array([[int(nodeI.x - width / 2), nodeI.y],
[int(nodeI.x - width / 2), nodeJ.y]])
pts2 = np.array([[int(nodeI.x + width / 2), nodeI.y],
[int(nodeI.x + width / 2), nodeJ.y]])
pts1 = transformSetOfPointsAndReturn(pts1, H)
pts2 = transformSetOfPointsAndReturn(pts2, H)
cv2.line(self.grid, (pts1[0][0], pts1[0][1]),
(pts1[1][0], pts1[1][1]), (255, 0, 0), 1)
cv2.line(self.grid, (pts2[0][0], pts2[0][1]),
(pts2[1][0], pts2[1][1]), (255, 0, 0), 1)
def get5nearestNeighbours(self, vehicleIndex):
dist = {}
a = np.array([
self.vehicles[vehicleIndex].curX, self.vehicles[vehicleIndex].curY
])
for i, vehicle in enumerate(self.vehicles):
if (i == vehicleIndex) or (len(vehicle.track)) < 2 or (
vehicle.track[-2][0] == -1 and vehicle.track[-2][1] == -1):
continue
b = np.array([vehicle.curX, vehicle.curY])
d = np.linalg.norm(a - b)
if len(dist.keys()) < 5:
dist[d] = i
continue
mx = max(dist.keys())
if mx > d:
dist.pop(mx, None)
return dist
def updateExportVehiclesIndices(self):
l = []
for i in self.exportVehicleIndices:
v1 = []
print i,
if i == -1 or len(self.vehicles[i].track) < 2 or (
self.vehicles[i].track[-2][0] == -1
and self.vehicles[i].track[-2][1] == -1):
v1 += [-1] * 17
else:
v1 += self.vehicleTypes.oneHotEncoding(self.vehicles[i].cl)
x = self.vehicles[i].track[-2][0]
y = self.vehicles[i].track[-2][1]
v1 += [x, y]
print v1
neighs = self.get5nearestNeighbours(i)
print neighs
for n in neighs.keys():
if self.vehicles[
neighs[n]].track[-2][0] != -1 and self.vehicles[
neighs[n]].track[-2][1] != -1:
v1 += [
x - self.vehicles[neighs[n]].track[-2][0],
y - self.vehicles[neighs[n]].track[-2][1]
]
else:
v1 += [-1, -1]
for k in range(len(neighs.keys()), 5):
v1 += [-1, -1]
if self.vehicles[i].track[-2][0] == -1 and self.vehicles[
i].track[-2][1] == -1:
v1 += [0, 0]
else:
v1 += [
self.vehicles[i].track[-2][0] - self.vehicles[i].curX,
self.vehicles[i].track[-2][1] - self.vehicles[i].curY
]
print v1,
l.append(v1)
print "\n",
for i, vehicle in enumerate(self.vehicles):
#print "Guru ", i, vehicle.numberOfEdgesCompleted
if vehicle.numberOfEdgesCompleted == 1 and i in self.exportVehicleIndices:
self.exportVehicleIndices[self.exportVehicleIndices.index(
i)] = -1
elif i not in self.exportVehicleIndices:
if -1 in self.exportVehicleIndices:
self.exportVehicleIndices[self.exportVehicleIndices.index(
-1)] = i
def drawVehicles(self):
for vehicle in self.vehicles:
if vehicle.curX == -1 and vehicle.curY == -1:
continue
vehi_size = self.vehicleTypes.getSize(vehicle.cl)
if self.supportVehicleSize:
print "Rectangles to be printed are : ", vehicle.curX - vehi_size[
0] / 2, vehicle.curY - vehi_size[
1] / 2, " ", vehicle.curX + vehi_size[
0] / 2, vehicle.curY + vehi_size[1] / 2
cv2.rectangle(self.grid, (vehicle.curX - vehi_size[0] / 2,
vehicle.curY - vehi_size[1] / 2),
(vehicle.curX + vehi_size[0] / 2,
vehicle.curY + vehi_size[1] / 2),
vehicle.color,
thickness=cv2.cv.CV_FILLED)
#cv2.putText(self.grid, str(vehicle.identity), (vehicle.curX-vehi_size[0]/2,vehicle.curY-vehi_size[1]/2), cv2.FONT_HERSHEY_SIMPLEX, 0.75, (255,0,0), 1, cv2.CV_AA)
else:
self.grid[vehicle.curY, vehicle.curX] = [255, 0, 0]
def drawVehiclesWithHomography(self, H):
for vehicle in self.vehicles:
if vehicle.curX == -1 and vehicle.curY == -1:
continue
vehi_size = self.vehicleTypes.getSize(vehicle.cl)
if self.supportVehicleSize:
pts = np.array([[
vehicle.curX - vehi_size[0] / 2,
vehicle.curY - vehi_size[1] / 2
],
[
vehicle.curX + vehi_size[0] / 2,
vehicle.curY - vehi_size[1] / 2
],
[
vehicle.curX + vehi_size[0] / 2,
vehicle.curY + vehi_size[1] / 2
],
[
vehicle.curX - vehi_size[0] / 2,
vehicle.curY + vehi_size[1] / 2
]],
dtype=float)
pts = transformPoly(pts, H)
self.grid = drawPoly(self.grid, pts, vehicle.color)
#cv2.rectangle(self.grid,(vehicle.curX-vehi_size[0]/2,vehicle.curY-vehi_size[1]/2), (vehicle.curX+vehi_size[0]/2,vehicle.curY+vehi_size[1]/2), (0,255,0), 1)
else:
self.grid[vehicle.curY, vehicle.curX] = [255, 0, 0]
def updateGrid(self):
self.grid = np.zeros(self.size)
self.drawNetwork()
self.drawVehicles()
def updateGridWithHomography(self, H, perspective_road_points,
straight_road_points):
self.grid = np.zeros(self.size)
#self.grid = drawPoly(self.grid, perspective_road_points)
#self.grid = drawPoly(self.grid, straight_road_points)
self.drawNetworkWithHomography(H)
self.drawVehiclesWithHomography(H)
def simulateAndVisualize(self):
t = 0
while (t < self.totalTime):
print "=========== Frame no : ", t, " =========="
#for vehicle in self.vehicles:
# vehicle.move(t, self.grid, self.vehicles)
self.moveall(t)
#self.moveallNearestNeighbour(t)
self.updateGrid()
t += 1
cv2.imshow("sim", self.grid)
if cv2.waitKey(33) == 27:
break
#if random.random() < 0.1:
# self.vehicles.append(self.addVehicle())
# cv2.waitKey(0)
cv2.destroyAllWindows()
def simulateAndVisualizeWithHomography(self, H, perspective_road_points,
straight_road_points):
t = 0
video = "/Users/gramaguru/Desktop/car_videos/sing.mp4"
cap = cv2.VideoCapture(video)
while (t < self.totalTime):
ret, frame = cap.read()
print frame.shape
frame = misc.imresize(frame, (384, 512, 3))
for vehicle in self.vehicles:
vehicle.move(t, self.grid, self.vehicles)
self.updateGridWithHomography(H, perspective_road_points,
straight_road_points)
t += 1
cv2.imshow("orginal", frame)
cv2.imshow("sim", self.grid)
if cv2.waitKey(33) == 27:
break
#if random.random() < 0.1:
# self.vehicles.append(self.addVehicle())
# cv2.waitKey(0)
cv2.destroyAllWindows()
def simulateContinuousTraffic(self,
total_time,
speed=False,
lanes=False,
vehicleTypes=False):
t = 0
while (t < total_time):
for vehicle in self.vehicles:
vehicle.move(t, self.grid, self.vehicles)
self.updateGrid()
#cv2.imshow("sim", self.grid)
#if cv2.waitKey(33) == 27:
# break
#if t % 5 == 0:
if random.random() <= 0.2:
if speed:
self.vehicles.append(self.addVehicleWithSpeed())
else:
self.vehicles.append(self.addVehicle(lanes, vehicleTypes))
#self.vehicleSequences.append(self.vehicleTypes.oneHotEncoding(vehicle.cl) + self.roadNetwork.edges[vehicle.numberOfEdgesCompleted].oneHotEncoding(vehicle.laneNo))
if speed:
self.vehicleSequences.append(vehicle.cl + " " +
str(vehicle.laneNo) + " " +
str(vehicle.speed))
else:
self.vehicleSequences.append(vehicle.cl + " " +
str(vehicle.laneNo))
#print "Guru ", self.vehicleTypes.oneHotEncoding(vehicle.cl) + self.roadNetwork.edges[vehicle.numberOfEdgesCompleted].oneHotEncoding(vehicle.laneNo)
else:
if speed:
self.vehicleSequences.append(("None 0 0"))
else:
self.vehicleSequences.append(("None 0"))
#self.updateExportVehiclesIndices()
t += 1
#print self.vehicleSequences
self.exportGeneratedVehicles(speed)
cv2.destroyAllWindows()
def generateTrafficUsingNN(self, total_time, start_seq):
# Parse command line arguments
t = 0
#decoder = torch.load("/Users/gramaguru/ComputerVision/computer-vision/Traffic/IndianTraffic/trafficGenerator/trafficNextPrediction/exportGeneratedAlphabets.pt")
#decoder = torch.load("/Users/gramaguru/ComputerVision/computer-vision/Traffic/IndianTraffic/trafficGenerator/trafficNextPrediction/traffic_seq_road_only_cars.pt")
decoder = torch.load(
"/Users/gramaguru/ComputerVision/computer-vision/Traffic/IndianTraffic/trafficGenerator/trafficNextPrediction/traffic_seq_roadNetwork_speed.export_51.pt"
)
parameter_len = 3
seq = generate_couple(decoder, start_seq, total_time, 0.8, False,
parameter_len)
print seq
while (t < total_time):
v = seq[parameter_len * t:parameter_len * t + parameter_len]
laneNo = int(v[1])
if parameter_len == 2:
speed = 4
elif parameter_len == 3:
speed = int(v[2])
if v[0] != 'n' and speed != 0:
self.vehicles.append(
Vehicle(self.vehicleTypes.map[v[0]], self.roadNetwork, 0,
SimpleLaneMovement(), speed, laneNo))
for vehicle in self.vehicles:
vehicle.move(t, self.grid, self.vehicles)
self.updateGrid()
cv2.imshow("NNsim", self.grid)
if cv2.waitKey(33) == 27:
break
t += 1
def simulateAndExport(self):
t = 0
while (t < self.totalTime):
for vehicle in self.vehicles:
vehicle.move(t, self.grid, self.vehicles)
self.updateGrid()
t += 1
self.export()
def exportGeneratedVehicles(self, speed=False):
#exportGeneratedSeq = raw_input()
if speed:
exportGeneratedSeq = "traffic_seq_prob_dist_speed.txt"
else:
exportGeneratedSeq = "traffic_seq_prob_dist.txt"
print self.vehicleSequences
fp = open(exportGeneratedSeq, "w")
fp.write(str(self.vehicleSequences))
fp.close()
def export(self):
exportFileName = "traffic_" + str(len(
self.roadNetwork.edges)) + "_" + str(len(
self.vehicles)) + "_" + str(time.time()) + ".txt"
fp = open(exportFileName, "w")
trafficSummary = str(self.grid.shape[0]) + " " + str(
self.grid.shape[1]) + " " + str(self.totalTime) + "\n"
fp.write(trafficSummary)
fp.write(str(len(self.roadNetwork.edges)) + "\n")
for edge in self.roadNetwork.edges:
fp.write(
str(edge.node1.x) + " " + str(edge.node1.y) + " " +
str(edge.node2.x) + " " + str(edge.node2.y) + " " +
str(edge.width) + "\n")
fp.write(str(len(self.vehicles)) + "\n")
for vehicle in self.vehicles:
fp.write(vehicle.cl + "\n")
tracklen = len(vehicle.track)
diff = 0
if tracklen < self.totalTime:
diff = self.totalTime - tracklen
for t in range(self.totalTime):
if t < diff:
fp.write(str(t) + " " + str(-1) + " " + str(-1) + "\n")
continue
fp.write(
str(t) + " " + str(vehicle.track[t - diff][0]) + " " +
str(vehicle.track[t - diff][1]) + "\n")
fp.close()
def getSpeeds(self, vehicles):
speeds = {}
for vehicle in vehicles:
speeds[vehicle.identity] = vehicle.prev_velocity
return speeds
def get_vehicles_lanes(self, vehicles):
lanes_frames = {}
for vehicle in vehicles:
if vehicle.curX not in lanes_frames.keys():
lanes_frames[vehicle.curX] = [[
vehicle.identity, vehicle.curX, vehicle.curY
]]
else:
lanes_frames[vehicle.curX].append(
[vehicle.identity, vehicle.curX, vehicle.curY])
return lanes_frames
def get_vehicles_lanes_dist(self, vehicles, lanes):
lanes_frames = {}
for vehicle in vehicles:
lane = min(lanes, key=lambda x: abs(x - vehicle.curX))
if lane not in lanes_frames.keys():
lanes_frames[lane] = [[
vehicle.identity, vehicle.curX, vehicle.curY
]]
else:
lanes_frames[lane].append(
[vehicle.identity, vehicle.curX, vehicle.curY])
return lanes_frames
def getVehiclesInScreen(self, vehicles):
filtered_vehicles = []
for vehicle in vehicles:
if vehicle.curX == -1 or vehicle.curY == -1 or vehicle.numberOfEdgesCompleted != 0:
continue
filtered_vehicles.append(vehicle)
return filtered_vehicles
def getVehiclesToStart(self, timestamp):
veh = []
for vehicle in self.vehicles:
if vehicle.curX == -1 and vehicle.curY == -1 and vehicle.startTime <= timestamp and vehicle.numberOfEdgesCompleted == 0:
print "start this vehicle ", vehicle.startTime, vehicle.identity, timestamp, vehicle
veh.append(vehicle)
return veh
def updateRemainingVehicles(self, timestamp):
for vehicle in self.vehicles:
if vehicle.curX == -1 and vehicle.curY == -1 and vehicle.startTime > timestamp:
vehicle.track.append((-1, -1))
if vehicle.numberOfEdgesCompleted != 0:
#vehicle.track.append((-1,-1))
vehicle.curX = -1
vehicle.curY = -1
def splitVehiclesAndUpdate(self, timestamp):
screen_veh = []
start_veh = []
for vehicle in self.vehicles:
if vehicle.curX == -1 and vehicle.curY == -1 and vehicle.startTime <= timestamp and vehicle.numberOfEdgesCompleted == 0:
start_veh.append(vehicle)
elif vehicle.curX == -1 and vehicle.curY == -1 and vehicle.startTime > timestamp:
vehicle.track.append((-1, -1))
elif vehicle.numberOfEdgesCompleted != 0:
vehicle.curX = -1
vehicle.curY = -1
elif vehicle.curX == -1 or vehicle.curY == -1 or vehicle.numberOfEdgesCompleted != 0:
continue
else:
screen_veh.append(vehicle)
return screen_veh, start_veh
def rectangleOverLap(self, ltp1x, ltp1y, rbp1x, rbp1y, ltp2x, ltp2y, rbp2x,
rbp2y):
if (ltp1x > rbp2x or ltp2x > rbp1x):
return False
if (ltp1y > rbp2y or ltp2y > rbp1y):
return False
return True
def positionempty(self, posX, posY, curVehicle, vehicles):
vehi_x = curVehicle.curX
vehi_y = curVehicle.curY
#print curVehicle.cl
curVehicleSize = self.vehicleTypes.getSize(curVehicle.cl)
rec1ltx = posX - curVehicleSize[0] / 2
rect1lty = posY - curVehicleSize[1] / 2
rect1rbx = posX + curVehicleSize[0] / 2
rect1rby = posY + curVehicleSize[1] / 2
for vehicle in vehicles:
if vehi_x == vehicle.curX and vehi_y == vehicle.curY:
continue
vehi_size = self.vehicleTypes.getSize(vehicle.cl)
rect2ltx = vehicle.curX - vehi_size[0] / 2
rect2lty = vehicle.curY - vehi_size[1] / 2
rect2rbx = vehicle.curX + vehi_size[0] / 2
rect2rby = vehicle.curY + vehi_size[1] / 2
if self.rectangleOverLap(rec1ltx, rect1lty, rect1rbx, rect1rby,
rect2ltx, rect2lty, rect2rbx,
rect2rby):
#print "check this man " + str(vehicle.curX) + "," + str(vehicle.curY) + " " + str(posX) + " " + str(posY)
return False
return True
def moveall(self, timestamp):
# create feature vector for all the vehicles
route = self.roadNetwork
initNode = route.edges[0].node1
finalNode = route.edges[0].node2
width = route.edges[0].width
lanes = route.edges[0].lanes
roadLeftBottomX = initNode.x - width / 2
roadLeftBottomY = initNode.y
roadRightTopX = finalNode.x + width / 2
roadRightTopY = finalNode.y
laneszz = [
roadLeftBottomX + ((i * width) / lanes) + (width / lanes) / 2
for i in range(4)
]
print "Lanes are : ", laneszz
#vehicles = self.getVehiclesInScreen(self.vehicles)
#print "vehicles in the screen are ", vehicles
#vehicles_to_start = self.getVehiclesToStart(timestamp)
#print "vehicles to start are ", vehicles_to_start
#self.updateRemainingVehicles(timestamp)
vehicles, vehicles_to_start = self.splitVehiclesAndUpdate(timestamp)
print "vehicles in the screen are ", vehicles
print "vehicles to start are ", vehicles_to_start
speeds = self.getSpeeds(vehicles)
vehicles_per_lanes = self.get_vehicles_lanes_dist(vehicles, laneszz)
#print "speeds : ", speeds
#print "==="
print "vehicle per lanes : ", vehicles_per_lanes
#print "===="
## Start few vehicles.
for curVehicle in vehicles_to_start:
posX = curVehicle.curX
posY = curVehicle.curY
identity = curVehicle.identity
route = curVehicle.route
edgeNo = curVehicle.numberOfEdgesCompleted
speed = curVehicle.speed
laneNo = curVehicle.laneNo
shape = self.grid.shape
sizeX = shape[0]
sizeY = shape[1]
tr = 0
while 1:
x = laneNo
x = random.choice(laneszz)
if self.positionempty(x, roadLeftBottomY, curVehicle,
vehicles):
print "wth man " + str(x) + " " + str(roadLeftBottomY)
curVehicle.updatePos(edgeNo, x, roadLeftBottomY)
break
tr += 1
if tr >= 100:
curVehicle.updatePos(edgeNo, posX, posY)
break
feature_vecs = []
## Move the already present vehicles.
for curVehicle in vehicles:
vehicle = np.array([
curVehicle.identity, curVehicle.curX, curVehicle.curY,
self.dataLoader.carType[curVehicle.cl]
])
feature_vec = np.array(
self.dataLoader.getFeatureVectorImp(vehicle, speeds,
vehicles_per_lanes,
laneszz))
#feature_vec.resize(1, feature_vec.shape[0])
print feature_vec
#feature_vec = (feature_vec - self.min_xval)/(self.max_xval - self.min_xval)
feature_vecs.append(feature_vec)
pred_vels = np.array([])
print "feature vecs are ", feature_vecs
if len(feature_vecs) != 0:
feature_vecs = np.array(feature_vecs)
pred_vels = self.loaded_model.predict(feature_vecs)
#pred_vels = pred_vels*(self.max_yval-self.min_yval) + self.min_yval
# update all the vehicles current positions etc.
fp = open("velocity_log", "a")
for i, (vel, vehicle) in enumerate(zip(pred_vels, vehicles)):
posX = vehicle.curX
posY = vehicle.curY
edgeNo = vehicle.numberOfEdgesCompleted
shape = self.grid.shape
sizeX = shape[0]
sizeY = shape[1]
vehicle.prev_velocity = [0, vel[0]]
fp.write(str(vel[0]) + "\n")
posY -= int(vel)
print "updated vehicles ", vehicle, posY, vel[0]
if (posY > sizeX + 11):
vehicle.updatePos(-1, -1, -1)
elif posY < 0:
vehicle.updatePos(edgeNo + 1, -1, -1)
else:
vehicle.updatePos(edgeNo, posX, posY)
print "updated vehicles ", vehicles
print "started vehicles ", vehicles_to_start
fp.close()
def moveallNearestNeighbour(self, timestamp):
# create feature vector for all the vehicles
route = self.roadNetwork
initNode = route.edges[0].node1
finalNode = route.edges[0].node2
width = route.edges[0].width
lanes = route.edges[0].lanes
roadLeftBottomX = initNode.x - width / 2
roadLeftBottomY = initNode.y
roadRightTopX = finalNode.x + width / 2
roadRightTopY = finalNode.y
#laneszz = [roadLeftBottomX + ((i*width)/lanes) + (width/lanes)/2 for i in range(4)]
laneszz = [
roadLeftBottomX + ((i * width) / lanes) + (width / lanes) / 2
for i in range(2)
]
print "Lanes are : ", laneszz
vehicles, vehicles_to_start = self.splitVehiclesAndUpdate(timestamp)
print "vehicles in the screen are ", vehicles
print "vehicles to start are ", vehicles_to_start
speeds = self.getSpeeds(vehicles)
#print "speeds : ", speeds
#print "==="
#print "vehicle per lanes : ", vehicles_per_lanes
#print "===="
## Start few vehicles.
for curVehicle in vehicles_to_start:
posX = curVehicle.curX
posY = curVehicle.curY
identity = curVehicle.identity
route = curVehicle.route
edgeNo = curVehicle.numberOfEdgesCompleted
speed = curVehicle.speed
laneNo = curVehicle.laneNo
shape = self.grid.shape
sizeX = shape[0]
sizeY = shape[1]
tr = 0
while 1:
x = laneNo
x = random.choice(laneszz)
if self.positionempty(x, roadLeftBottomY, curVehicle,
vehicles):
print "wth man " + str(x) + " " + str(roadLeftBottomY)
curVehicle.updatePos(edgeNo, x, roadLeftBottomY)
break
tr += 1
if tr >= 100:
curVehicle.updatePos(edgeNo, posX, posY)
break
feature_vecs = []
pred_vels = []
## Move the already present vehicles.
vehs = []
for curVehicle in vehicles:
vehicle = np.array([
curVehicle.identity, curVehicle.curX, curVehicle.curY,
self.dataLoader.carType[curVehicle.cl]
])
vehs.append(vehicle)
vehs = np.array(vehs)
feature_vecs, _ = self.dataLoader1.getNNFeaturesFromFrame(
vehs, speeds, None, None, True)
if len(feature_vecs) != 0:
feature_vecs = (feature_vecs - self.min_xval1) / (self.max_xval1 -
self.min_xval1)
feature_vecs = np.array(feature_vecs)
pred_vels = self.loaded_model1.predict(feature_vecs)
pred_vels = pred_vels * (self.max_yval1 -
self.min_yval1) + self.min_yval1
# update all the vehicles current positions etc.
for i, (vel, vehicle) in enumerate(zip(pred_vels, vehicles)):
posX = vehicle.curX
posY = vehicle.curY
edgeNo = vehicle.numberOfEdgesCompleted
shape = self.grid.shape
sizeX = shape[0]
sizeY = shape[1]
vehicle.prev_velocity = [vel[0], vel[1]]
posY -= (vel[1])
posX += (vel[0])
print "updated vehicles ", vehicle, posY, vel[0]
if (posY > sizeX + 11):
vehicle.updatePos(-1, -1, -1)
elif posY < 0:
vehicle.updatePos(edgeNo + 1, -1, -1)
else:
vehicle.updatePos(edgeNo, posX, posY)
print "updated vehicles ", vehicles
print "started vehicles ", vehicles_to_start