Class none person, how to remove it ?

[Rasantis]

Search before asking

• I have searched the Supervision issues and found no similar feature requests.

Question

I'm running this code on my raspberry pi 4 with picamera to detect and count people, using yolov8n, but sometimes it detects a class called none person, and then shows several boxes, when crossing the line it ends up counting these boxes with the class none person, and then, it ends up uncalibrating the count... I didn't find it in the documentation about this class called none person... how to disable it?

`import cv2
import json
import numpy as np
from picamera2 import Picamera2
from ultralytics import YOLO
import supervision as sv
import os

class PiLineCounter:
def init(self, lines_json_path, model_path):
with open(lines_json_path, 'r') as f:
self.lines_data = json.load(f)
self.model = YOLO(model_path)

    # Inicialização dos anotadores
    self.line_annotator = sv.LineZoneAnnotator(
        thickness=1,
        text_thickness=1,
        text_scale=1,
        custom_in_text="entrando",
        custom_out_text="saindo"
    )
    
    self.box_annotator = sv.BoxAnnotator(
        thickness=2,
        text_thickness=1,
        text_scale=0.5
    )

    # Inicialização da PiCamera2
    self.picam2 = Picamera2()
    preview_config = self.picam2.create_preview_configuration()
    self.picam2.configure(preview_config)
    self.picam2.start()

    # Inicialização dos contadores de linha
    self.line_counters = {}
    self.initialize_counters()

def initialize_counters(self):
    for line_key, line_value in self.lines_data.items():
        # Usar as coordenadas das linhas diretamente do JSON
        start_point_x, start_point_y = line_value['points'][0]
        end_point_x, end_point_y = line_value['points'][1]
        start_point = sv.Point(start_point_x, start_point_y)
        end_point = sv.Point(end_point_x, end_point_y)
        self.line_counters[line_key] = sv.LineZone(start=start_point, end=end_point)

def run(self):
    while True:
            frame = self.picam2.capture_array()
            frame = cv2.cvtColor(np.array(frame), cv2.COLOR_RGB2BGR)

            results = self.model.track(frame, show=False, stream=False, agnostic_nms=True, imgsz=320)
            print(f"Número de resultados de detecção: {len(results)}")
            for result in results:
                detections = sv.Detections.from_ultralytics(result)

                if detections is None or len(detections.xyxy) == 0:
                    print("Nenhuma detecção neste frame. Pulando...")
                    continue

                # Imprimir todas as detecções e seus respectivos class_id, labels e confianças
                for d in detections:
                    class_id = d[3]
                    label = self.model.model.names[class_id]
                    confidence = d[2]
                    print(f"Detecção: class_id={class_id}, label={label}, confiança={confidence:.2f}")

                detections_filtered = [d for d in detections if d[3] == 0]
                print(f"Número de detecções de pessoas: {len(detections_filtered)}")


                labels = [f"{d[4]} {self.model.model.names[d[3]]} {d[2]:0.2f}" for d in detections_filtered]

                for line_key in self.line_counters.keys():
                    in_count, out_count = self.line_counters[line_key].trigger(detections=detections_filtered)
                    print(f"Linha {line_key}: Entrando - {in_count}, Saindo - {out_count}")

                # Criar um objeto Detections que possa ser usado pelo BoxAnnotator
                if len(detections_filtered) > 0:
                    xyxy = np.array([d[0] for d in detections_filtered])
                    confidences = np.array([d[2] for d in detections_filtered])
                    class_ids = np.array([d[3] for d in detections_filtered])
                    tracker_ids = np.array([d[4] for d in detections_filtered])

                    detections_for_annotation = sv.Detections(xyxy=xyxy, confidence=confidences, class_id=class_ids, tracker_id=tracker_ids)

                    frame = self.box_annotator.annotate(
                        scene=frame,
                        detections=detections_for_annotation,
                        labels=labels
                    )
                else:
                    print("Nenhuma detecção de pessoas neste frame.")

                for line_key in self.line_counters.keys():
                    self.line_annotator.annotate(frame=frame, line_counter=self.line_counters[line_key])

            # Exibir o frame original sem redimensionamento
            cv2.imshow("PiCamera Line Counter", frame)
            #cv2.imwrite('/dev/shm/frame.jpg', frame)

            if cv2.waitKey(1) & 0xFF == ord('q'):
                break
    cv2.destroyAllWindows()

if name == 'main':
lines_json_path = "lines_with_doubled_data.json"

model_path = "yolov8n.pt"

pi_line_counter = PiLineCounter(
    lines_json_path=lines_json_path,
    model_path=model_path
)
pi_line_counter.run()

`

Additional

lines_with_doubled_data.json

[SkalskiP]

Hi, @Rasantis 👋🏻 ! Thanks a lot for your interest in supervision. I am sorry that I have not been responsive for the last few days. Before Christmas, I was busy with duties unrelated to supervision, and I was off for the last few days.

Before we start, let's convert this issue into a discussion and move it to the Q&A section.

[SkalskiP]

@Rasantis

It looks like the None is the tracker_id value associated with the given detection. Probably YOLOv8 can't assign the correct tracker_id so it leaves it as None.

You can:

1. Skip tracker_id when printing the labels. You can do that by updating:

# old line
labels = [f"{d[4]} {self.model.model.names[d[3]]} {d[2]:0.2f}" for d in detections_filtered]

# new line
labels = [f"{self.model.model.names[d[3]]} {d[2]:0.2f}" for d in detections_filtered]

2. Filter out detections with tracker_id equal to None.

detections_filtered = detections[detections.tracker_id != None]

[Rasantis]

it worked, thanks a lot @SkalskiP !

[SkalskiP]

that's grat to hear!