Skip to content

Commit

Permalink
Changes in the traffic light detection to work with the new AI model.
Browse files Browse the repository at this point in the history
  • Loading branch information
@Glas42
Glas42 committed Jun 19, 2024
1 parent 621903b commit f5930c3
Show file tree
Hide file tree
Showing 5 changed files with 117 additions and 89 deletions.
3 changes: 3 additions & 0 deletions blog/V1.11.Minor.md
View file
Original file line number Diff line number Diff line change
Expand Up @@ -23,6 +23,9 @@ title: Version 1.11 > Minor
This changelog will contain all changes from 1.11.0 onwards before the next major version.
!!!

==- Version 1.11.43
Changes in the traffic light detection to work with the new AI model.
[!badge variant="dark" text="Glas42"] [!badge variant="ghost" text="TrafficLightDetection"]
==- Version 1.11.42
Fixed problems with Navigation Detection AI, Default Steering and Cruise Control.
[!badge variant="dark" text="Glas42"] [!badge variant="ghost" text="NavigationDetection, Default Steering, Cruise Control"]
Expand Down
2 changes: 1 addition & 1 deletion changelog.txt
View file
Original file line number Diff line number Diff line change
@@ -1 +1 @@
Fixed problems with Navigation Detection AI, Default Steering and Cruise Control.
Changes in the traffic light detection to work with the new AI model.
4 changes: 2 additions & 2 deletions plugins/NavigationDetection/main.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -284,7 +284,7 @@ def LoadAIModelThread():
CheckForAIModelUpdates()
while AIModelUpdateThread.is_alive(): time.sleep(0.1)

if GetAIModelName() == []:
if GetAIModelName() == "UNKNOWN":
return

LoadAIProgress = 0
Expand Down Expand Up @@ -498,7 +498,7 @@ def GetAIModelProperties():


if UseAI:
if TorchAvailable == True:
if TorchAvailable == True and settings.GetSettings("NavigationDetection", "UseAI", False) == True:
helpers.RunInMainThread(LoadAIModel)
else:
print("NavigationDetectionAI not available due to missing dependencies.")
Expand Down
195 changes: 110 additions & 85 deletions plugins/TrafficLightDetection/main.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -383,33 +383,42 @@ def get_text_size(text="NONE", text_width=100, max_text_height=100):


def ClassifyImage(image):
global IMG_WIDTH
global IMG_HEIGHT
try:
if AIModelUpdateThread.is_alive(): return True
if AIModelLoadThread.is_alive(): return True
except:
return True
try:
image = cv2.resize(image, (IMG_WIDTH, IMG_HEIGHT))
except:
IMG_WIDTH = GetAIModelProperties()[2]
IMG_HEIGHT = GetAIModelProperties()[3]
if IMG_WIDTH == "UNKNOWN" or IMG_HEIGHT == "UNKNOWN":
print(f"TrafficLightDetection - Unable to read the AI model image size. Make sure you didn't change the model file name. The code wont run the TrafficLightDetectionAI.")
console.RestoreConsole()
return True
image = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
transform = transforms.Compose([
transforms.ToTensor(),
])
image = transform(image).unsqueeze(0).to(AIDevice)

image = np.array(image, dtype=np.float32)
if IMG_CHANNELS == 'Grayscale' or IMG_CHANNELS == 'Binarize':
image = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
else:
image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
if IMG_CHANNELS == 'RG':
image = np.stack((image[:, :, 0], image[:, :, 1]), axis=2)
elif IMG_CHANNELS == 'GB':
image = np.stack((image[:, :, 1], image[:, :, 2]), axis=2)
elif IMG_CHANNELS == 'RB':
image = np.stack((image[:, :, 0], image[:, :, 2]), axis=2)
elif IMG_CHANNELS == 'R':
image = image[:, :, 0]
image = np.expand_dims(image, axis=2)
elif IMG_CHANNELS == 'G':
image = image[:, :, 1]
image = np.expand_dims(image, axis=2)
elif IMG_CHANNELS == 'B':
image = image[:, :, 2]
image = np.expand_dims(image, axis=2)
image = cv2.resize(image, (IMG_WIDTH, IMG_HEIGHT))
image = image / 255.0
if IMG_CHANNELS == 'Binarize':
image = cv2.threshold(image, 0.5, 1.0, cv2.THRESH_BINARY)[1]

image = transforms.ToTensor()(image).unsqueeze(0).to(AIDevice)
with torch.no_grad():
output = np.array(AIModel(image)[0].tolist())
probabilities = np.exp(output - np.max(output)) / np.sum(np.exp(output - np.max(output)))
confidence = np.max(probabilities)
predicted_class = np.argmax(probabilities)
return True if predicted_class != 3 else False
obj_class = np.argmax(output)
return True if obj_class != 3 else False


def HandleCorruptedAIModel():
Expand All @@ -432,22 +441,19 @@ def LoadAIModelThread():
global LoadAIProgress
global AIModel
global AIModelLoaded
global IMG_WIDTH
global IMG_HEIGHT

CheckForAIModelUpdates()
while AIModelUpdateThread.is_alive(): time.sleep(0.1)

if GetAIModelName() == []:
if GetAIModelName() == "UNKNOWN":
return

LoadAIProgress = 0
LoadAILabel = "Loading the AI model..."

print("\033[92m" + f"Loading the AI model..." + "\033[0m")

IMG_WIDTH = GetAIModelProperties()[2]
IMG_HEIGHT = GetAIModelProperties()[3]
GetAIModelProperties()

ModelFileCorrupted = False

Expand Down Expand Up @@ -626,33 +632,54 @@ def DeleteAllAIModels():


def GetAIModelProperties():
global MODEL_METADATA
global IMG_WIDTH
global IMG_HEIGHT
global IMG_CHANNELS
global MODEL_EPOCHS
global MODEL_BATCH_SIZE
global MODEL_IMAGE_COUNT
global MODEL_TRAINING_TIME
global MODEL_TRAINING_DATE
try:
ModelFolderExists()
if GetAIModelName() == "UNKNOWN":
return ("UNKNOWN", "UNKNOWN", "UNKNOWN", "UNKNOWN", "UNKNOWN", "UNKNOWN", "UNKNOWN")
else:
string = str(GetAIModelName())
if string != "UNKNOWN":
epochs = int(string.split("EPOCHS-")[1].split("_")[0])
batch = int(string.split("BATCH-")[1].split("_")[0])
img_width = int(string.split("IMG_WIDTH-")[1].split("_")[0])
img_height = int(string.split("IMG_HEIGHT-")[1].split("_")[0])
img_count = int(string.split("IMG_COUNT-")[1].split("_")[0])
training_time = string.split("TIME-")[1].split("_")[0]
training_date = string.split("DATE-")[1].split(".")[0]
return (epochs, batch, img_width, img_height, img_count, training_time, training_date)
else:
return ("UNKNOWN", "UNKNOWN", "UNKNOWN", "UNKNOWN", "UNKNOWN", "UNKNOWN", "UNKNOWN")
MODEL_METADATA = {"data": []}
IMG_WIDTH = "UNKNOWN"
IMG_HEIGHT = "UNKNOWN"
IMG_CHANNELS = "UNKNOWN"
MODEL_EPOCHS = "UNKNOWN"
MODEL_BATCH_SIZE = "UNKNOWN"
MODEL_IMAGE_COUNT = "UNKNOWN"
MODEL_TRAINING_TIME = "UNKNOWN"
MODEL_TRAINING_DATE = "UNKNOWN"
torch.jit.load(os.path.join(f"{variables.PATH}plugins/TrafficLightDetection/AIModel", GetAIModelName()), _extra_files=MODEL_METADATA, map_location=AIDevice)
MODEL_METADATA = str(MODEL_METADATA["data"]).replace('b"(', '').replace(')"', '').replace("'", "").split(", ")
for var in MODEL_METADATA:
if "image_width" in var:
IMG_WIDTH = int(var.split("#")[1])
if "image_height" in var:
IMG_HEIGHT = int(var.split("#")[1])
if "image_channels" in var:
IMG_CHANNELS = str(var.split("#")[1])
if "epochs" in var:
MODEL_EPOCHS = int(var.split("#")[1])
if "batch" in var:
MODEL_BATCH_SIZE = int(var.split("#")[1])
if "image_count" in var:
MODEL_IMAGE_COUNT = int(var.split("#")[1])
if "training_time" in var:
MODEL_TRAINING_TIME = var.split("#")[1]
if "training_date" in var:
MODEL_TRAINING_DATE = var.split("#")[1]
except Exception as ex:
exc = traceback.format_exc()
SendCrashReport("TrafficLightDetection - Error in function GetAIModelProperties.", str(exc))
print(f"TrafficLightDetection - Error in function GetAIModelProperties: {ex}")
console.RestoreConsole()
return ("UNKNOWN", "UNKNOWN", "UNKNOWN", "UNKNOWN", "UNKNOWN", "UNKNOWN", "UNKNOWN")


if UseAI:
if TorchAvailable == True:
if TorchAvailable == True and settings.GetSettings("TrafficLightDetection", "UseAI", False) == True:
helpers.RunInMainThread(LoadAIModel)
else:
print("TrafficLightDetectionAI not available due to missing dependencies.")
Expand Down Expand Up @@ -1141,47 +1168,6 @@ def plugin(data):
coordinates.append((round(x + w * 0.5), round(offset), w, h, colorstr))



##################################################################################################
# Start: Code to send traffic light images to drive if enabled
##################################################################################################
if send_traffic_light_images == True:
try:
if last_traffic_light_image + 0.5 < time.time() and len(coordinates) > 0:
for x, y, w, h, state in coordinates:
tld_y1 = round(y1 + y - h*4)
if tld_y1 < 0:
tld_y1 = 0
elif tld_y1 > frameFull.shape[0]:
tld_y1 = frameFull.shape[0]
tld_y2 = round(y1 + y + h * 4)
if tld_y2 < 0:
tld_y2 = 0
elif tld_y2 > frameFull.shape[0]:
tld_y2 = frameFull.shape[0]
tld_x1 = round(x1 + x - w * 2.5)
if tld_x1 < 0:
tld_x1 = 0
elif tld_x1 > frameFull.shape[1]:
tld_x1 = frameFull.shape[1]
tld_x2 = round(x1 + x + w * 2.5)
if tld_x2 < 0:
tld_x2 = 0
elif tld_x2 > frameFull.shape[1]:
tld_x2 = frameFull.shape[1]
traffic_light_image = frameFull[tld_y1:tld_y2, tld_x1:tld_x2].copy()
threading.Thread(target=SendImage, args=(traffic_light_image, round(x1 - tld_x1 + x), round(y1 - tld_y1 + y), w, h,), daemon=True).start()
last_traffic_light_image = time.time()
except Exception as e:
exc = traceback.format_exc()
SendCrashReport("TrafficLightDetection - TrafficLightDetectionAI data collection error.", str(exc))
print("TrafficLightDetection - TrafficLightDetectionAI data collection error: " + str(exc))
##################################################################################################
# end: Code to send traffic light images to drive if enabled
##################################################################################################



try:
# Tracking with IDs:
def generate_new_id():
Expand Down Expand Up @@ -1249,6 +1235,45 @@ def generate_new_id():
approved = ClassifyImage(image_classification)
else:
approved = True

##################################################################################################
# Start: Code to send traffic light images to drive if enabled
##################################################################################################
if send_traffic_light_images == True:
try:
if last_traffic_light_image + 0.5 < time.time() and len(coordinates) > 0:
for x, y, w, h, state in coordinates:
tld_y1 = round(y1 + y - h*4)
if tld_y1 < 0:
tld_y1 = 0
elif tld_y1 > frameFull.shape[0]:
tld_y1 = frameFull.shape[0]
tld_y2 = round(y1 + y + h * 4)
if tld_y2 < 0:
tld_y2 = 0
elif tld_y2 > frameFull.shape[0]:
tld_y2 = frameFull.shape[0]
tld_x1 = round(x1 + x - w * 2.5)
if tld_x1 < 0:
tld_x1 = 0
elif tld_x1 > frameFull.shape[1]:
tld_x1 = frameFull.shape[1]
tld_x2 = round(x1 + x + w * 2.5)
if tld_x2 < 0:
tld_x2 = 0
elif tld_x2 > frameFull.shape[1]:
tld_x2 = frameFull.shape[1]
traffic_light_image = frameFull[tld_y1:tld_y2, tld_x1:tld_x2].copy()
threading.Thread(target=SendImage, args=(traffic_light_image, round(x1 - tld_x1 + x), round(y1 - tld_y1 + y), w, h,), daemon=True).start()
last_traffic_light_image = time.time()
except Exception as e:
exc = traceback.format_exc()
SendCrashReport("TrafficLightDetection - TrafficLightDetectionAI data collection error.", str(exc))
print("TrafficLightDetection - TrafficLightDetectionAI data collection error: " + str(exc))
##################################################################################################
# end: Code to send traffic light images to drive if enabled
##################################################################################################

trafficlights.append((nearestpoint, ((None, None, None), (head_x, head_z, angle, head_rotation_degrees_x), (head_x, head_z, angle, head_rotation_degrees_x)), new_id, approved))

# Remove lost traffic lights from the list, the traffic light which has the highest distance to the nearest traffic light in the current frame gets removed
Expand Down Expand Up @@ -1733,9 +1758,9 @@ def OpenWiki():
def InstallCUDAPopup():
helpers.Dialog("Warning: CUDA is only available for NVIDIA GPUs!", f"1. Check on https://wikipedia.org/wiki/CUDA#GPUs_supported which CUDA version your GPU supports.\n2. Go to https://pytorch.org/ and copy the download command for the corresponding CUDA version which is compatible with your GPU.\n (Select Stable, Windows, Pip, Python and the CUDA version you need)\n3. Open your file explorer and go to {os.path.dirname(os.path.dirname(variables.PATH))} and run the activate.bat\n4. Run this command in the terminal which opened after running the activate.bat: 'pip uninstall torch torchvision torchaudio'\n5. After the previous command finished, run the command you copied from the PyTorch website and wait for the installation to finish.\n6. Restart the app and the app should automatically detect CUDA as available and use your GPU for the AI.", ["Exit"], "Exit")
helpers.MakeButton(trackeraiFrame, "Install CUDA for GPU support", InstallCUDAPopup, 5, 0, width=30, sticky="nw")
model_properties = GetAIModelProperties()
GetAIModelProperties()
helpers.MakeLabel(trackeraiFrame, "Model properties:", 6, 0, font=("Robot", 12, "bold"), sticky="nw")
helpers.MakeLabel(trackeraiFrame, f"Epochs: {model_properties[0]}\nBatch Size: {model_properties[1]}\nImage Width: {model_properties[2]}\nImage Height: {model_properties[3]}\nImages/Data Points: {model_properties[4]}\nTraining Time: {model_properties[5]}\nTraining Date: {model_properties[6]}", 7, 0, sticky="nw")
helpers.MakeLabel(trackeraiFrame, f"Epochs: {MODEL_EPOCHS}\nBatch Size: {MODEL_BATCH_SIZE}\nImage Width: {IMG_WIDTH}\nImage Height: {IMG_HEIGHT}\nImages/Data Points: {MODEL_IMAGE_COUNT}\nTraining Time: {MODEL_TRAINING_TIME}\nTraining Date: {MODEL_TRAINING_DATE}", 7, 0, sticky="nw")
self.progresslabel = helpers.MakeLabel(trackeraiFrame, "", 9, 0, sticky="nw")
self.progress = ttk.Progressbar(trackeraiFrame, orient="horizontal", length=238, mode="determinate")
self.progress.grid(row=10, column=0, sticky="nw", padx=5, pady=0)
Expand Down
2 changes: 1 addition & 1 deletion version.txt
View file
Original file line number Diff line number Diff line change
@@ -1 +1 @@
1.11.42
1.11.43

0 comments on commit f5930c3

Please sign in to comment.