Traffic Sign Detection using YOLOv11

• Traffic Sign Detection using YOLOv11
  • Data
  • Model
  • Fine Tuning
    • YOLO11n summary (fused)
    • Run summary
  • Detections
  • Dependencies
  • Project Setup
  • Limitations
  • Conclusion
  • Acknowledgements

Data

The Self-Driving Cars Dataset is used to train the traffic sign detection model. It contains 4969 total images split into train, val and test sets with 3530, 801 and 638 images of dimension 416x416 respectively. The dataset contains images of 15 different traffic signs.

The classes available in the dataset are:

1. all
2. Green Light
3. Red Light
4. Speed Limit 100
5. Speed Limit 110
6. Speed Limit 120
7. Speed Limit 20
8. Speed Limit 30
9. Speed Limit 40
10. Speed Limit 50
11. Speed Limit 60
12. Speed Limit 70
13. Speed Limit 80
14. Speed Limit 90
15. Stop

Model

The yolo11n version of the model is used to fine-tune on the dataset. The model was trained for 50 epochs with batch size 16.

Note: The .ipy notebook is not uploaded due to privacy issues.

Fine Tuning

YOLO11n summary (fused)

238 layers, 2,585,077 parameters, 0 gradients, 6.3 GFLOPs

Class	Images	Instances	Box(P	R	mAP50	mAP50-95)
all	801	944	0.95	0.905	0.959	0.836
Green Light	87	122	0.901	0.743	0.851	0.525
Red Light	74	108	0.891	0.722	0.844	0.529
Speed Limit 100	52	52	0.95	0.942	0.989	0.889
Speed Limit 110	17	17	0.916	1	0.986	0.915
Speed Limit 120	60	60	1	0.943	0.995	0.908
Speed Limit 20	56	56	0.981	0.93	0.985	0.871
Speed Limit 30	71	74	0.963	0.959	0.984	0.924
Speed Limit 40	53	55	0.935	0.945	0.988	0.887
Speed Limit 50	68	71	0.973	0.915	0.98	0.886
Speed Limit 60	76	76	0.92	0.912	0.96	0.89
Speed Limit 70	78	78	0.987	0.962	0.981	0.9
Speed Limit 80	56	56	0.96	0.929	0.973	0.866
Speed Limit 90	38	38	0.954	0.789	0.924	0.784
Stop	81	81	0.975	0.982	0.988	0.929

Run summary

Parameter / Metric	Value
lr/pg0	2e-05
lr/pg1	2e-05
lr/pg2	2e-05
metrics/mAP50(B)	0.95912
metrics/mAP50-95(B)	0.83597
metrics/precision(B)	0.95049
metrics/recall(B)	0.90534
model/GFLOPs	6.456
model/parameters	2592765
model/speed_PyTorch(ms)	3.062
train/box_loss	0.47508
train/cls_loss	0.33472
train/dfl_loss	0.90102
val/box_loss	0.55826
val/cls_loss	0.34385
val/dfl_loss	0.95067

Detections

Dependencies

• python 3.x
• opencv_contrib_python
• opencv_python
• ultralytics

Project Setup

1. Make a virtual environment using the following command:

   python3 -m venv myenv

   Replace myenv with the name you want for your virtual environment. This will create a folder named myenv in your current directory containing the virtual environment files.

2. Activate the virtual environment:

   source myenv/bin/activate

   Remember to replace myenv with the actual name of the environment created in the previous step.

3. Clone the repository:

   git clone https://github.com/bhaskrr/traffic-sign-detection-using-yolov11.git

4. Navigate to the root directory of the project:

   cd path/to/the/project

5. Install dependencies:

   pip install -r requirements.txt

6. Run the scripts

   • To process images

     python3 process_image.py

   • To process videos

     python3 process_video.py

Limitations

The model tends to perform better with images and is fairly accurate with videos in certain scenarios. So, to obtain better results with videos or real-time data, consider fine-tuning a deeper model available on ultralytics docs.

Conclusion

This project demonstrates how a fine-tuned YOLOv11 model can be used for traffic sign detection.

Here are a few use cases for this project:

1. Autonomous Vehicle Navigation: The model can be used in self-driving car systems to recognize traffic signs accurately. This would enable autonomous vehicles to follow traffic rules and regulations, analyzing every sign whether it’s about speed limit or stop-and-go indications to navigate the roads safely.

2. Traffic Rule Compliance: This model can be used in driver assistance systems to ensure that drivers comply with all traffic rules. Alerts can be generated when drivers exceed the speed limit or don't stop at red lights, fostering safer roads.

3. Road Safety Training Programs: This model allows Driving schools and automotive companies to build simulations and education programs. These programs can guide new drivers in identifying and responding to different traffic signs, thus enhancing road safety knowledge.

4. Smart City Infrastructure: City authorities could use this model in connected CCTV or IoT infrastructure to track and monitor traffic compliance in real time, helping identify areas with frequent rule violations for potential improvement.

5. Road Network Analysis: Transportation engineering researchers can use this model to analyze how efficiently different sign classes are distributed and recognized around the city. This data can be instrumental in planning more efficient and safer road networks.

Acknowledgements

The media files used to test the model predictions are taken from pexels.com.