• Download 13
• File Size 415.51 KB
• File Count 1
• Create Date 16/03/2025
• Last Updated 16/03/2025

Road Terrain Detection Using Computer Vision and M.L

Road Terrain Detection Using Computer Vision and M.L

 

 

Mr. S. K. Mahajan

Department of Computer

Technology

K. K. Wagh Polytechnic,

Nashik, India

skmahajan@kkwagh.edu.in

Mrunal B. Derale

Department of Computer

Technology

K. K. Wagh Polytechnic,

Nashik, India

mrunalderale@gmail.com

Rushikesh S. Jadhav

Department of Computer                                                                                   Technology

K. K. Wagh Polytechnic,

Nashik, India

jrushi725@gmail.com

Kaveri R. Deshmukh

Department of Computer

Technology

K. K. Wagh Polytechnic

Nashik, India deshmukhhkaveri@gmail.com

Piyush S. Dighe

Department of Computer Technology

K. K. Wagh Polytechnic,

Nashik, India

piyushdighe777@gmail.com

        Abstract :Terrain detection is a critical component in autonomous navigation systems, enabling vehicles to adapt their operational parameters based on the terrain type for safety and efficiency. This paper presents the design and implementation of a deep learning-based terrain detection system, which utilizes a Convolutional Neural Network (CNN) model trained on a diverse dataset of terrain images. The proposed system achieves real-time inference and accurately classifies terrains, such as asphalt, gravel, and grass, using video feed from an onboard camera. The dataset used for training comprises high-resolution images, preprocessed to a uniform size of 100x100 pixels, ensuring consistency in input dimensions.

     The system is trained on a TensorFlow 2.x framework, leveraging its optimized computational graph and SavedModel format for deployment. Transfer learning techniques were employed to enhance performance on limited datasets, reducing computational overhead while maintaining high accuracy. The inference engine outputs classifications in a user-friendly GUI designed using Tkinter, displaying the detected terrain type and corresponding action suggestions.

     The results demonstrate the system's robustness, achieving high classification accuracy even under varying lighting conditions. This work underscores the potential of deep learning in terrain detection and sets the stage for further integration into advanced driver-assistance systems (ADAS). Future extensions include expanding the dataset to cover more complex terrains and integrating the detection system with vehicle control mechanisms for real-time response.

Keywords: Terrain Detection, Convolutional Neural Networks, Autonomous Navigation, Deep Learning, TensorFlow, Real-time Classification