fr/modes/track/

[giscus[bot]]

fr/modes/track/

Apprends à utiliser Ultralytics YOLO pour le suivi d'objets dans les flux vidéo. Guides pour utiliser différents trackers et personnaliser les configurations des trackers.

https://docs.ultralytics.com/fr/modes/track/

[Hamzaboughfir123]

I would like to count the number of all detections after tracking

[pderrenger]

Hey there! 🌟 To count the total number of detections after tracking, you can use the len() function on the .boxes attribute of your results object. Here's a quick example:

from ultralytics import YOLO

# Load your model and perform tracking
model = YOLO('yolov8n.pt')  # Adjust the model path as needed
results = model.track(source='path/to/your/video.mp4')

# Count the number of detections
num_detections = len(results.boxes)
print(f"Total detections: {num_detections}")

This counts and prints the total number of detections after tracking. Hope this helps and happy tracking! 😊

[ghost]

Hi there !

Please I have some questions.

1. How to initialize the detector or tracking with these parameters ( ) = (0 1 877 427 419 510) ?

2. How to get these parameters for each frame during the tracking and save it in to a file ?

3. I specify that object of interest during the tracking is only the "cup". By reading documentation, I solved this way, but the two others questions No.

[ghost]

So after the detection I call now model.track(), where and how ? I don't undersatand

[pderrenger]

Hey there! 🌞

For initializing and updating the tracker in your code, you can directly use the model.track() method after loading your model. The tracker will handle the frame updates internally. Here's a quick example on how to do it:

from ultralytics import YOLO

# Load your YOLOv8 model
model = YOLO('path/to/your/model.pt')

# To start tracking, just call the .track() method with your source video.
# The tracker will initialize and update automatically as it processes the video frames.
results = model.track(source='path/to/video.mp4', show=True)  # shows the tracking in real-time

# You can specify the tracker type if you have a preference, like ByteTrack.
# results = model.track(source='path/to/video.mp4', tracker='bytetrack.yaml', show=True)

To save tracking data (like frame id, object id, bbox coordinates), you can iterate over the results and write them to a file:

with open('tracking_results.txt', 'w') as file:
    for frame_id, result in enumerate(results):
        for det in result.pred[0]:
            xmin, ymin, xmax, ymax, conf, cls = det[:6]  # Extract bbox details
            file.write(f"{frame_id} {cls} {xmin} {ymin} {xmax-xmin} {ymax-ymin} {conf}\n")

Note: Adjust the variable extraction based on how the Results object stores detections in your version of Ultralytics YOLO.

If you're running into the 'Results' object has no attribute 'pred', make sure you're using the latest version of the Ultralytics YOLO package, or access the detections using attributes specific to your installed version, usually through .xyxy[0] for bbox details.

Hope this helps get you up and running with tracking! 😊 If you have any more questions or need further assistance, feel free to ask!

[ghost]

Hello

I would like to evaluate (apply) the ultralytics yolov8 tracking with (on) MOT17 or MOT20. Do you know how to proceed ?

[ghost]

And then apply The HOTA metric

[pderrenger]

@Aris115 hey there! 👋

Great to hear you're interested in applying Ultralytics YOLOv8 tracking on MOT17 or MOT20 datasets! Here's a quick way to get started:

1. Prepare your Data: Ensure your MOT17 or MOT20 datasets are in a format recognized by YOLOv8 (Typically, this means having images/videos with corresponding annotations).

2. Load and Run YOLOv8 Tracking: Using the Ultralytics YOLOv8 pre-trained model or your own trained model, run the tracking as follows:

from ultralytics import YOLO

# Load your model
model = YOLO('path/to/your/model.pt')

# Track on a MOT17 or MOT20 video
results = model.track(source='path/to/MOT17/or/MOT20/video.mp4')

3. Evaluate Using HOTA: Currently, you might need an external tool to evaluate the tracking results with HOTA, as direct support might not be detailed in the documentation. HOTA evaluation can be done with available tools like TrackEval.

For precise steps, especially on evaluation with HOTA, you might want to explore external resources or forums related to MOT challenges or consider contributing to the Ultralytics docs if you find a novel approach! 💡

Let us know if you have any more questions or if there's anything else we can help with. Happy tracking! 🚀

[syariffortask]

fr/modes/track/

Apprends à utiliser Ultralytics YOLO pour le suivi d'objets dans les flux vidéo. Guides pour utiliser différents trackers et personnaliser les configurations des trackers.

https://docs.ultralytics.com/fr/modes/track/

hello can you help me how to get id_track and class_id for each detection,and put id_track and class_id/classname in above of bounding box?

from ultralytics import YOLO
import cv2

model = YOLO("yolov8n.pt")
cap = cv2.VideoCapture(0)

assert cap.isOpened(),"error membaca kamera"

region_points = [(20, 400), (1080, 404), (1080, 360), (20, 360)]

while cap.isOpened():
ret,frame = cap.read()
if not ret:
break
trackss = model.track(frame,persist=True,classes=[0],verbose=False)
for track in trackss:
class_id = track.boxes.cls
cls_name = model.names[class_id]
trak_id = track.boxes.id
print(trak_id)
rr=track.boxes.xyxy
print(rr)
for r in rr:
# x1,x2,y1,y2=map(int,r)

  x1 = int(r[0])
  y1 = int(r[1])
  x2 = int(r[2])
  y2 = int(r[3])
  center_x = int(x1+x2)//2
  center_y = int(y1+y2)//2
  
  cv2.rectangle(frame,(x1,y1),(x2,y2),(0,255,0),2)
  cv2.putText(frame,str(cls_name),(x1,y1-10),cv2.FONT_HERSHEY_SIMPLEX,0.5,(0,255,0),2)
  cv2.putText(frame,str(track.boxes.id),(x1,y1+10),cv2.FONT_HERSHEY_SIMPLEX,0.5,(0,255,0),2)
  cv2.circle(frame, (center_x, center_y), 5, (0, 0, 255), -1)

cv2.imshow("track",frame)
if cv2.waitKey(1) & 0xFF==27:
break

cap.release()
cv2.destroyAllWindows()

[ghost]

Hey!

Where can I find the link (article) that describes YoloV8 architecture. Thanks!

[glenn-jocher]

Hey there!

You're asking about a fantastic resource! 🌟 For detailed information on the YOLOv8 architecture, I'd recommend diving into the official Ultralytics documentation. While we don't have a dedicated article specifically outlining the YOLOv8 architecture yet, our docs and GitHub repo are packed with insights and updates that will help you understand the model's structure and improvements.

For the deeper technical nuances of YOLOv8, examining the source code on our GitHub repo might also offer valuable clarity. We often update our docs and resources, so keep an eye out for new information!

Hope this helps, and happy coding! 😊

[ghost]

Hi! I need your help.

I'm trying to evaluate tracking after using ultralytics YoloV8 with TrackEval using MOT17's gt.txt file and the tracking results I did res.txt on MOT17-05-FRCNN images.

Thank you.

[ghost]

With HOTA, DetA, AssA, LocA and HOTA(0) metrics

[glenn-jocher]

Hey there!

Thanks for reaching out. 🤗 For evaluating YOLOv8 with TrackEval, it sounds like you've got everything pretty well set up using the gt.txt file from MOT17 and your res.txt results.

As for the "x", "y", and "z" values, those aren't directly provided by YOLOv8 tracking outputs since it primarily focuses on 2D object detection and tracking. These coordinates typically represent 3D spatial info which might require additional data or methods to derive.

However, if you're all set and just looking to move forward with TrackEval, ensure your res.txt file follows the MOT Challenge result format closely. It should work smoothly with TrackEval as long as the formats align!

Wishing you the best with your tracking evaluation. If you have more questions or need further assistance, feel free to ask! 😊

[Daniel-lucifer]

Personnel occlusion and changes in lighting angle can affect the detection results. How can I improve YOLOv8 to address these issues

[glenn-jocher]

Hey there! 👋

For improving YOLOv8 detection performance under varying conditions like occlusion and lighting changes, here are a couple of suggestions:

1. Data Augmentation: Ensure your training dataset includes examples of occlusions and different lighting scenarios. This helps the model learn to recognize objects under these conditions. You can augment your dataset with images that simulate these conditions or specifically collect such data.

2. Fine-tuning: If you're already using a pre-trained YOLOv8 model, consider fine-tuning it on a dataset that represents your specific use case, including occlusions and varied lighting.

Here's an example of how you might fine-tune a model with custom data:

from ultralytics import YOLO

# Assuming you have a custom dataset prepared
model = YOLO('yolov8n.pt')  # Load pre-trained model

# Fine-tune the model
results = model.train(data='custom_data.yaml', epochs=50, imgsz=640)

Remember, the key is diversity in your training data! The more scenarios your model sees during training, the better it can generalize to new conditions. 🚀

[Daniel-lucifer]

If the accuracy of localization is affected by personnel occlusion, can yolov8 be used to train a dataset related to people, remove the parts recognized as people, and not participate in the subsequent feature point detection and matching work

[glenn-jocher]

@Daniel-lucifer hey there! 👋

Yes, you can definitely use YOLOv8 to train on a dataset focused on people. Afterward, you can identify parts recognized as people and exclude them from subsequent feature point detection and matching tasks. 🚀

For excluding detected people, you could manually process the output bounding boxes from the model predictions and filter out those identified as people. Here is a small code snippet to give you an idea:

from ultralytics import YOLO

# Load the model
model = YOLO('yolov8n.pt')  # or your custom trained model

# Predict on your image
results = model('your_image.jpg')

# Filter out detections
for det in results.pred[0]:
    if results.names[int(det[-1])] == 'person':
        continue  # Skip person detections
    # Process non-person detections

This way, you focus only on the features that are not identified as people. Hope this helps! Let us know how it goes. 😊

[Hamzaboughfir123]

i would like to register the output video that show the process of tracking with the number of all object detected while tracking process

[glenn-jocher]

Bonjour,

Pour associer des données GPS à chaque objet détecté selon le temps exact de détection, vous pouvez synchroniser les métadonnées de temps de la vidéo avec vos données GPS. Voici comment vous pourriez procéder :

1. Enregistrement du Temps de Détection : Lors de la détection d'un objet, enregistrez le moment exact de la détection en utilisant le numéro de l'image et le framerate de la vidéo pour calculer le temps en secondes.

import cv2
from ultralytics import YOLO

# Charger le modèle
model = YOLO('path/to/model.pt')

# Ouvrir la vidéo
cap = cv2.VideoCapture('path/to/video.mp4')
fps = cap.get(cv2.CAP_PROP_FPS)

detections = []

while cap.isOpened():
    ret, frame = cap.read()
    if not ret:
        break

    frame_number = cap.get(cv2.CAP_PROP_POS_FRAMES)
    results = model.track(frame, persist=True)
    
    if len(results[0].boxes) > 0:
        time_detected = frame_number / fps
        detections.append((time_detected, results[0]))

# Ici, `detections` contient des tuples de (temps, résultats)

2. Corrélation avec les Données GPS : Vous pouvez désormais associer les temps de détection à vos données GPS.

gps_data = {...}  # Vos données GPS avec des timestamps

for detection_time, result in detections:
    corresponding_gps = find_closest_timestamp(detection_time, gps_data)
    print(f"Objet détecté à {detection_time}s avec GPS {corresponding_gps}")

Cette méthode vous permet de lier précisément les temps de détection d'objets aux données GPS correspondantes, en fonction des horodatages. Assurez-vous que les horodatages entre votre système vidéo et les données GPS sont bien synchronisés ou ajustez-les en conséquence.

Bonne continuation avec votre projet de tracking! 😊

[Hamzaboughfir123]

merci mais au lieu de récupérer le temps de détection je veux récupérer le temps ou un objet est identifié non seulement détecte et retourner le temps ou chaque objet est identifié(lorsqu'il a un id unique)

[glenn-jocher]

Bonjour,

Pour enregistrer le temps précis où chaque objet est identifié avec un ID unique, vous pouvez capturer le temps la première fois qu'un ID spécifique apparaît dans le suivi. Voici un exemple simple utilisant Python :

import cv2
from ultralytics import YOLO

# Charger le modèle
model = YOLO('yolov8n.pt')

# Ouvrir la vidéo
cap = cv2.VideoCapture('path/to/video.mp4')
fps = cap.get(cv2.CAP_PROP_FPS)

# Dictionnaire pour enregistrer le temps de première identification
first_identified = {}

while cap.isOpened():
    ret, frame = cap.read()
    if not ret:
        break

    frame_number = cap.get(cv2.CAP_PROP_POS_FRAMES)
    results = model.track(frame, persist=True)
    
    for i, box in enumerate(results[0].boxes):
        obj_id = int(box.id)
        if obj_id not in first_identified:
            first_identified[obj_id] = frame_number / fps
            print(f"Objet ID {obj_id} identifié à {first_identified[obj_id]:.2f} secondes")

# Affichage de l'heure d'identification pour tous les objets suivis
for obj_id, time in first_identified.items():
    print(f"Objet ID {obj_id} a été identifié pour la première fois à {time:.2f} secondes")

Ce script suit les objets au fil du temps et enregistre le premier moment où chaque nouvel ID est détecté. N'oubliez pas de remplacer 'path/to/video.mp4' et 'yolov8n.pt' avec vos chemins appropriés. J'espère que cela vous aidera à obtenir l'association temporelle que vous recherchez!

[Hamzaboughfir123]

merci c'est tres utile.

[glenn-jocher]

Merci beaucoup pour votre retour, je suis ravi que vous trouviez cela utile ! Si vous avez d'autres questions ou besoin d'assistance supplémentaire en cours de route, n'hésitez pas à demander. Bonne continuation avec votre projet de suivi d'objets ! 😊