Traffic signal voilation detection from an IP camera using AI

Project Name: Traffic signal violation detection from an IP camera using AI

This project aims to develop a system that uses neural networks for detecting and recognising the number plates of vehicles on the road, which violates traffic rules using the BeagleBoard AI64 hardware. This uses an IP camera to stream real-time video data at 1920x1080 at 30 fps, which is fed to the custom model trained for recognising the number plates of the vehicle based on the Inception-v1 model.

The goal of the project is to automate the traffic signal violation detection and recognition system and make it easy for the traffic police department to monitor real-time traffic and take action against the violating vehicle owner in a fast and efficient way. The major task of the project is to train the CNN to detect the violation and recognise the number plates of the vehicle.

TI SDK to train the CNN based on the Inception v1 model and to build a C++ application for the Beaglebone AI64 target.

Approach and timeline summary:

  1. Understanding the TI SDK stack (1 week)
  2. Dataset collection and preparation (2 weeks)
  3. Training the model (2 weeks)
  4. Developing the C++ application and testing (2 weeks)
  5. Testing and re-training if required based on the (2 week buffer time
  6. Analysis of the performance (1 week)
  7. Report Generation (0.5 Week)

Total Size: ~300 hours


  1. CNN must be trained to recognise the number plates.
  2. Real-time video data must be streamed from the IP camera, and traffic rule violations must be detected.

In the highly trafficked area, number plates will be hidden by other vehicles. Hence, it is difficult to develop a very highly accurate system to recognise the violation of the traffic rules. A possible solution is to have cameras in multiple locations. An algorithm must be implemented to process video data from multiple cameras to avoid overlap problems.

Further improvements:
Cloud-based technologies shall be introduced to save records in the database and make them accessible anytime over the network.

1 Like