# detection.py

CLASS_DISPLAY_NAMES = {

"Police Car": "Police Car",

"Arrow Board": "Arrow Board",

"Traffic Cones": "Traffic Cones",

"Accident": "Accident",

}

def detect_events(model, image, conf_thresh=0.7, device="cpu", half=False, imgsz=640):

"""

model: ultralytics.YOLO

image: PIL.Image or ndarray

device: "cuda" | "mps" | "cpu"

half: bool (True for CUDA FP16, False otherwise)

"""

# Call Ultralytics predict with explicit device/precision

# NOTE: half=True is effective only on CUDA

results = model(

image,

conf=conf_thresh,

device=device,

half=half,

imgsz=imgsz

)

detections = []

r0 = results[0]

for box in r0.boxes:

conf = float(box.conf)

cls_idx = int(box.cls)

cls_name = model.names[cls_idx]

if conf >= conf_thresh:

# xyxy is a (1,4) tensor → take [0] and cast to float

x1, y1, x2, y2 = [float(v) for v in box.xyxy[0].tolist()]

detections.append({

"class": cls_name,

"displayName": CLASS_DISPLAY_NAMES.get(cls_name, cls_name),

"confidence": conf,

"bbox": {"x1": x1, "y1": y1, "x2": x2, "y2": y2},

})

return detections