## Cyber Security of Image for Self-driving Cars
**2022 GiCS**
**Github:** https://github.com/jschen9999/2022GiCS

### Introduction
With the continuous innovation in autonomous driving, the number of automated vehicles is steadily increasing. To ensure the safe operation of autonomous vehicles, it is crucial not only to have sensors functioning properly and accurately but also to implement systems that prevent network attacks. However, as technology advances, attack methods become more sophisticated, and artificial intelligence-driven cyber-attacks pose new threats.
### Purpose
Therefore, we propose a solution to prevent malicious deepfake attacks on computer vision systems within the sensors of autonomous vehicles.
### Model Structure & Explanation
The computer vision system integrates cameras and artificial intelligence models. After capturing images with the camera, a classifier is set up before the image is input into the object detection model. The classifier is a convolutional neural network based on VGG-16. It is trained using a dataset comprising 22,000 images of real vehicle scenes and 22,000 images of Deepfake-generated attack scenes. The goal is to train the classifier to distinguish whether an image is real or a Deepfake attack.

Based on the classification results:
- If the result indicates a real vehicle image, the computer vision system continues normal operation.
- If it is identified as a Deepfake attack, an alternative computer vision system is activated.

This approach ensures that, in the event of a Deepfake attack on the computer vision system of autonomous vehicles, the system can promptly obtain the correct real image to continue safe operation.

### Training Data & Performance