Car Logo Detection Model
Team Members
Ravikumar Yadav, Krishna Singh, Omkar Upadhyay, Vinit Singh
Use Case Importance
Detecting car logos in real-time using computer vision can play an important role in intelligent transportation systems, automated parking management, vehicle identification, and smart surveillance applications. Car logo detection helps in recognizing vehicle brands quickly without requiring manual inspection.
In modern smart cities and automated monitoring systems, identifying vehicle manufacturers can assist in:
traffic analysis,
parking management,
security monitoring,
dealership analytics,
and automated vehicle record systems.
Manual identification of car brands from CCTV footage or parking surveillance is time-consuming and often inaccurate. Our solution automates this process using Artificial Intelligence and real-time object detection, enabling fast and reliable car logo recognition using a mobile camera or webcam feed.
Data Collection and Annotation
We manually collected a dataset of 500+ real-world car images from parking areas, roadsides, college campuses, and public vehicle spaces using mobile phone cameras. Images were captured under different practical conditions such as:
varying lighting conditions,
multiple camera angles,
different distances,
partial logo visibility,
motion blur,
and crowded backgrounds
to ensure that the model generalizes effectively in real-world scenarios.
The dataset included multiple car brands such as:
Toyota
BMW
Audi
Mercedes-Benz
All images were manually annotated using LabelMe, where bounding boxes were carefully drawn around each visible car logo to generate accurate ground truth labels.
After annotation, several augmentation techniques were applied to improve dataset diversity, including:
horizontal flipping,
brightness adjustment,
rotation,
zooming,
blur,
and contrast enhancement.
These augmentation techniques expanded the dataset to 1500+ training images.
The final dataset distribution was as follows:
780 Training Images
390 Validation Images
130 Test Images
This dataset preparation helped improve the robustness and generalization capability of the YOLOv9 model.
Model Training and Validation
The YOLOv9m object detection model was trained on Google Colab using a Tesla T4 GPU for efficient training performance.
During training, the following performance metrics were monitored:
Precision
Recall
mAP@0.5
mAP@0.5–0.95
Additionally, loss metrics such as:
Box Loss
Classification Loss
DFL Loss
were continuously analyzed to track model learning progress.
Several visualization outputs were used during validation, including:
Confusion Matrix
Precision-Recall Curve
F1-Score Curve
Training Loss Graphs
The generated outputs were stored in:
confusion_matrix.png
Results Achieved
Metric Result
Precision 97%
Recall 94%
mAP@50 95%
mAP@50-95 81%
The obtained results demonstrated that the model was capable of accurately detecting and classifying car logos under practical real-world conditions.
Model Deployment and Demo
The trained YOLOv9 model was successfully deployed on the YOLOvX mobile application by Wiserli, enabling accurate real-time car logo detection with satisfactory inference speed and stable FPS performance.
The deployed system demonstrated smooth real-time performance and reliable logo detection capability in practical testing environments.
Conclusion
The developed Car Logo Detection System performed successfully during both web-based and real-time mobile testing. The model achieved high precision and recall while detecting multiple car logos under different environmental conditions.
During experimentation, it was observed that:
lighting conditions,
logo size,
partial visibility,
and extreme viewing angles
slightly affected detection performance in some cases. However, the use of real-world custom images significantly improved the model’s practical generalization capability.
One of the major learnings from this project was the importance of:
dataset diversity,
accurate annotation,
and proper augmentation techniques
in improving object detection performance.
This project provided valuable hands-on experience in:
Computer Vision,
Object Detection,
YOLO-based model training,
Real-time deployment,
and practical AI system development.
Finally, we would like to thank Dr. Chandrakant Bothe Sir for conducting this internship workshop and providing guidance throughout the project, which helped us gain practical exposure to real-world AI implementation and deployment.