README

Introduction

This repository contains two deep learning models based on the ResNet-18 architecture, specifically designed for intelligent vehicle applications. The models enable an autonomous vehicle to perform two key tasks: line following and obstacle avoidance. These models are trained to process visual input from a camera and make real-time decisions to navigate efficiently and safely.

Models

1. Line Following Model

Architecture: ResNet-18
Input: RGB images (captured from a front-facing camera)
Output: Steering angle or direction commands to follow a predefined line/path
Training Data: Images of various road conditions with labeled steering angles
Usage: The model analyzes the input image and predicts the optimal steering angle to keep the vehicle on the desired path. It is trained to recognize and follow lines, even in different lighting and environmental conditions.

2. Obstacle Avoidance Model

Architecture: ResNet-18
Input: RGB images (captured from a front-facing camera)
Output: Decision commands (e.g., turn left, turn right, stop) to avoid obstacles
Training Data: Images with obstacles and corresponding avoidance actions
Usage: The model detects obstacles in the vehicle's path and provides real-time instructions to avoid collisions. It can handle various types of obstacles, such as walls, objects, and uneven terrain.

Dataset

The models are trained on custom datasets collected from real-world scenarios. The datasets include:

Line Following Dataset: Images of roads with labeled steering angles.
Obstacle Avoidance Dataset: Images with obstacles and corresponding avoidance actions.

The datasets are preprocessed to ensure robustness and generalization across different environments.

Real-Time Integration

To integrate these models into your intelligent vehicle system, you can use the following steps:

Capture real-time images from the vehicle's camera.
Preprocess the images using the AutoFeatureExtractor.
Run inference using the loaded models.
Send the predicted commands (steering angles or avoidance actions) to the vehicle's control system.

Training

if you need to retrain or fine-tune the models, you can use the Caffe framework along with the provided prototxt training configuration files.

License

This repository is licensed under the MIT License. You are free to use, modify, and distribute the code and models for any purpose.

Acknowledgments

ResNet-18: The models are based on the ResNet-18 architecture, which is a powerful and widely-used convolutional neural network.
Hugging Face Transformers: This repository utilizes the Hugging Face Transformers library for easy model deployment and inference.

MianXu
/

road_following_model