Zone-based counting

Count detections inside polygonal zones—ideal for traffic, retail, and other analytics.

Estimated read time: 4 minutes

Many applications (e.g., traffic, retail, safety) need counts within specific regions, not across the entire frame. Zone-based counting lets you define polygonal areas and track totals per zone, optionally per class. This guide uses a YOLOv8 detection model and an interactive zone tool so you can adjust zones live, without changing the code.

How it works

• A detection model runs on each frame to identify objects.
• A ZoneCounter analyzer checks which detections fall inside your polygons and updates the counts.
• You can interactively adjust zone vertices in the display window to fine-tune boundaries—no code changes needed.

Basic Zone Counter

Cars on a freeway passing through two zone counters.

Use this static setup when you want fixed polygon zones with automatic overlays. Update the coordinates in polygon_zones to match your scene so ZoneCounter can draw per-zone totals and populate inference_result.zone_counts.

Example

from degirum_tools import (
    ModelSpec,
    Display,
    ZoneCounter,
    AnchorPoint,
    remote_assets,
)
import degirum_tools

class_list = ["car", "motorbike", "truck"]
model_spec = ModelSpec(
    model_name="yolov8n_coco--640x640_quant_axelera_metis_1",
    zoo_url="degirum/axelera",
    inference_host_address="@local",
    model_properties={
        "device_type": ["AXELERA/METIS"],
        "overlay_color": [(255, 0, 0)],
        "output_class_set": set(class_list),
    },
)
model = model_spec.load_model()

video_source = remote_assets.traffic

polygon_zones = [
    [[265, 260], [730, 260], [870, 450], [120, 450]],
    [[400, 100], [610, 100], [690, 200], [320, 200]],
]

zone_counter = ZoneCounter(
    count_polygons=polygon_zones,
    class_list=class_list,
    per_class_display=True,
    triggering_position=AnchorPoint.CENTER,
)

degirum_tools.attach_analyzers(model, [zone_counter])

with Display("AI Camera") as output_display:
    for inference_result in degirum_tools.predict_stream(model, video_source):
        output_display.show(inference_result.image_overlay)
        print("Zone counts:", inference_result.zone_counts)

Interactive Zone Counter

Cars on a freeway passing through two interactive zone counters.

Pick the interactive workflow when you need to reposition zones during a live session. Drag an entire polygon with the left mouse button, then right-click a corner handle to refine that vertex while the stream keeps running.

• Interactive editing: click and drag zone vertices in the display window to reshape; press q to close the window.
• Classes: use output_class_set to filter which classes are counted; omit it to include all detected labels.
• Trigger point: AnchorPoint.CENTER works well for people and vehicles; try BOTTOM_CENTER for more ground-contact accuracy.
• Coordinates: define polygons in image pixels for the resolution you'll display.
• Performance: fewer, larger zones are more efficient than many small ones; filtering to needed classes also helps reduce overhead.
• Mouse controls: use the left mouse button (LMB) to drag the entire zone; right-click (RMB) a corner handle to adjust that vertex precisely.

Example

from degirum_tools import (
    ModelSpec,
    Display,
    ZoneCounter,
    AnchorPoint,
    remote_assets,
)
import degirum_tools

class_list = ["car", "motorbike", "truck"]
model_spec = ModelSpec(
    model_name="yolov8n_coco--640x640_quant_axelera_metis_1",
    zoo_url="degirum/axelera",
    inference_host_address="@local",
    model_properties={
        "device_type": ["AXELERA/METIS"],
        "overlay_color": [(255, 0, 0)],
        "output_class_set": set(class_list),
    },
)
model = model_spec.load_model()

video_source = remote_assets.traffic  # or your own video path

polygon_zones = [
    [[265, 260], [730, 260], [870, 450], [120, 450]],
    [[400, 100], [610, 100], [690, 200], [320, 200]],
]

zone_counter = ZoneCounter(
    count_polygons=polygon_zones,
    class_list=class_list,
    per_class_display=True,
    triggering_position=AnchorPoint.CENTER,
    window_name="AI Camera",
)

degirum_tools.attach_analyzers(model, [zone_counter])

with Display("AI Camera") as output_display:
    for inference_result in degirum_tools.predict_stream(model, video_source):
        output_display.show(inference_result.image_overlay)