Omdena Jakarta - Smart Traffic System
Project Links:
Background
Traffic Management refers to the combination of measures that serve to preserve traffic capacity and improve the security, safety and reliability of the overall road transport system. These measures make use of ITS systems, services and projects in day-to-day operations that impact on road network performance. Proper traffic management can ensure that
- Traffic flows smoothly and efficiently.
- Roads are well maintained and safe for all users, including pedestrians and cyclists.
- Congestion, local pollution and accidents are minimized.
- Vehicles are within speed limits and heading in correct direction along lane.
Solution
In this project we aim to provide smart solutions to address problems associated with Traffic Management
- Vehicle Category Classification and Detection.
- Traffic Density Classification.
- Pothole Object Detection.
No | Description | Dataset | Kaggle | Weights |
---|---|---|---|---|
1 | Vehicle Category Classification and Detection | Link | Link | Link |
2 | Traffic Density Classification | Link | Link | Link |
3 | Pothole Object Detection | Link | Link | Link |
π Skills
Pytorch, Torchvision, Ultralytics, OpenCV, Numpy, Streamlit, Git
Directory Tree
βββ notebooks
β βββ real-time-pothole-detection-using-ssd.ipynb
| |ββ traffic-density-classification-using-efficientnet.ipynb
β βββ vehicle-category-object-detection-pytorch.ipynb
βββ webapp
β βββ images
β β βββ category_img
β β βββ pothole_img
β β βββ density_img
β β βββ home_page
β βββ pages
β β βββ Pothole_Detection.py
β β βββ Traffic_Density_Classification.py
β β βββ Vehicle_Category_Detection.py
β βββ videos
β β βββ raw_clip.mp4
β β βββ annotated_clip.mp4
β βββ weights
β β βββ pothole_model.pth
β β βββ traffic_density.pth
β β βββ vehicle_categorization.pth
β βββ Introduction.py
β βββ utils.py
β βββ requirements.txt
β βββ packages.txt
βββ README.md
βββ LICENSE
βββ .gitignore
Vehicle Category Classification and Detection
- Reducing vehicle speed can prevent road crash incidents as an average 1 km increase in vehicle speed can lead to 3 percent increase in road accidents.
- Detecting and Categorizing different types of vehicle in traffic flow can help in determining vehicles speed using optical flow and enforcing speed limits.
- Single Shot Detector architecture with mobileNet_v3 backone is utilized for categorizing different Vehicles type.
- The model achieved a mean average precision of 0.645 on test dataset @ iou=0.50.
Traffic Density Classification
- The primary goal of traffic management is to make the movement of goods and persons as efficient, orderly, and safe as possible.
- Traffic Density Classification will be instrumental in redirecting traffic from major roads during peak hours and making the overall movement be safer and more efficient.
- EfficientNet_b0 architecture is utilized for traffic density classification into Empty, Low, Medium, High, Traffic Jam categories.
- The model achieved a mean average accuracy of 0.93 on test dataset.
Pothole Object Detection
- The safety and efficiency of travel systems depend on how well roads are kept up.
- Pothole Object Detection will be helpful in early identification of pothole and lowering long-term repair costs.
- It will be critical for ensuring the safety of drivers and the overall efficiency of transportation infrastructure.
- Single Shot Detector with MobileNetv3 backbone is utilized for detecting potholes on the roads.
- The model achieved a mean average precision of 0.512 at 0.50 intersection over union.
Future Development
- Road Lane Instance Segmentation
- Plate number recognition
- Vehicle tracking for speed determination
Run Webapp Locally
Clone the project
git clone https://github.com/OmdenaAI/jakarta-indonesia-predicting-road-defects
Change to project directory
cd webapp
Create Virtaul Environment and install dependencies
py -m venv venv
venv/Scripts/activate
pip install -r requirements.txt
Run Locally
streamlit run Introduction.py
Contributions
Chapter Lead: Louis Jefferson Zhang
Technical Lead: Sudhanshu Rastogi
Collaborators:
- Mohamed Chahed
- Mansoor Baig
- Nishtha Bhattacharjee
Project Archives
Navigate to archive branch to look into some project development notebooks.