Locating Missing Vehicles with AI: A Breakthrough in Traffic Surveillance
Summary
Researchers from the University of Toronto and Northeastern University have developed a deep learning-based vehicle image search engine that can help locate missing vehicles. This AI system, known as the Vehicle Image Search Engine (VISE), uses a region-based fully convolutional network (RFCN) for object detection and a ResNet-50-based CNN model to extract vehicle features. VISE can narrow down the location of a vehicle by refining search results based on time, location of cameras, and travel distance from the initial location of the vehicle. This technology has the potential to revolutionize traffic surveillance and help solve crimes involving missing vehicles.
The Challenge of Locating Missing Vehicles
Locating missing vehicles is a daunting task for law enforcement agencies. Traditional methods of searching for vehicles can be time-consuming and often yield few results. The use of AI in traffic surveillance has the potential to revolutionize the way missing vehicles are located.
How VISE Works
VISE uses a combination of object detection and feature extraction to identify vehicles in images. The system is trained on a large dataset of images from traffic cameras, which are sampled every two minutes. The images and metadata are stored in a SQLite database and then sent to an object detection and feature extraction system running on TensorFlow.
The system uses a region-based fully convolutional network (RFCN) for object detection, which is optimized using GPUs for faster inference. The feature extraction is done using a ResNet-50-based CNN model, which is trained on labeled pairs of images of objects from traffic camera footage.
Refining Search Results
One of the key features of VISE is its ability to refine search results based on time, location of cameras, and travel distance from the initial location of the vehicle. This allows users to narrow down the location of a vehicle and increase the chances of finding it.
Potential Applications
VISE has the potential to be used in a variety of applications, including:
- Traffic Surveillance: VISE can be used to monitor traffic conditions and locate missing vehicles.
- Crime Solving: VISE can be used to help solve crimes involving missing vehicles.
- Intelligent Transportation Systems: VISE can be integrated into intelligent transportation systems to improve traffic management and reduce congestion.
Technical Details
VISE is trained on a large dataset of images from traffic cameras, which are sampled every two minutes. The images and metadata are stored in a SQLite database and then sent to an object detection and feature extraction system running on TensorFlow.
The system uses a region-based fully convolutional network (RFCN) for object detection, which is optimized using GPUs for faster inference. The feature extraction is done using a ResNet-50-based CNN model, which is trained on labeled pairs of images of objects from traffic camera footage.
Performance
VISE has shown promising results in locating missing vehicles. The system can process images in real-time and provide accurate results. The use of GPUs for inference has significantly improved the performance of the system.
Future Work
Future work on VISE includes improving the accuracy of the system and integrating it into existing traffic surveillance systems. The use of more advanced AI techniques, such as generative models, could also be explored to improve the performance of the system.
Tables
| Feature | Description |
|---|---|
| Object Detection | Region-based fully convolutional network (RFCN) |
| Feature Extraction | ResNet-50-based CNN model |
| Dataset | Publicly available traffic camera footage from the Toronto Open Data Portal |
| Inference | Optimized using GPUs for faster inference |
| Refining Search Results | Based on time, location of cameras, and travel distance from the initial location of the vehicle |
References
The research on VISE was conducted by researchers from the University of Toronto and Northeastern University. The system was trained on a large dataset of images from traffic cameras, which are sampled every two minutes. The images and metadata are stored in a SQLite database and then sent to an object detection and feature extraction system running on TensorFlow.
Conclusion
VISE is a breakthrough in traffic surveillance that has the potential to revolutionize the way missing vehicles are located. The system uses a combination of object detection and feature extraction to identify vehicles in images and can refine search results based on time, location of cameras, and travel distance from the initial location of the vehicle. VISE has the potential to be used in a variety of applications, including traffic surveillance, crime solving, and intelligent transportation systems.