Adversarial Attack Detector in Self-Driving Vehicle

Introduction

In this project, we design a system of obeject detection intending for self-driving car usage. The system is able to defense adversarial attack. The system can identify the inputs which are attacked and we use YOLO to verify the results. demo video

System Architecture

Overview of Demo

Considering some limitation of the development board, we demonstrate the system in two different scenario. The overview of each scenarios is shown below. For more information, please refer to the demo part of our demo video.

Scenario 1: Clean data captured by camera

Scenario 2: Attacked data loaded with model

HW/SW Setup

1. Install Edge Impulse CLI
2. Install ARC GNU ToolChain
3. Donwlaod and Setup SDK
4. Clone this repository to your local computer
5. Modify the ROOT_PATH in scenario_1/Makefile and scenario_2/Makefile
6. Connect the WE-I to the computer by USB cable

User manual

To use the system for scenario 1:

1. $ cd scnario_1
2. Place tflite file under checkpopints/ directory. Pretrained model file is available at https://drive.google.com/file/d/1OZqE7vJ8KH-Pt2yfjFHCvaPEqgmCetMN/view?usp=sharing
3. $ make
4. $ make flash
5. $ himax-flash-tool -f output_gnu.img
6. Press the "reset" button on the board
7. $ python3 arc_detect.py #receive image from the baord and run YOLO inference

To use the system for scenario 2:

1. $ cd scenario_2
2. Copy one of test_sample from test_samples/ to src/
3. $ make
4. $ make flash
5. $ himax-flash-tool -f output_gnu.img
6. screen /dev/ttyUSB0 115200 #see the result
7. Verify the result using YOLO model with the images in test_sample_images/