DEVELOP AN APPLICATION TO CLASSIFY AND PREDICT THE ENVIRONMENT USING DEEP LEARNING FOR AN AUTONOMOUS VEHICLE

Computer Science and Engineering

The proposed paper focuses on the development of an innovative application designed for environment detection and prediction utilizing deep learning techniques, tailored specifically for autonomous vehicles. The goal is to create a system that can interpret and predict the surrounding environment from images and videos without relying on external sensors. The system seamlessly integrates real-time analysis from two distinct sources: a webcam feed and a dash-cam feed. Advanced object detection techniques are employed to predict and classify objects in the vehicle's surroundings, providing crucial information for safe autonomous navigation. The application employs YOLOv4-tiny, a state of the art object detection algorithm, to identify and classify objects in the environment and OpenCV is integrated for video and image processing, aiding in frame extraction and preprocessing tasks. The core functionality of the system revolves around real-time vehicle detection and distance estimation. By employing a predefined set of neural network models, utilizing predefined distance constants and object dimensions, the system calculates the distance between the detected vehicles and the camera, ensuring precise spatial awareness. Visual cues are incorporated to highlight application logic processes user inputs, vehicles, and potential hazards are flagged with an alert mechanism if they breach predefined safety thresholds, providing critical information for autonomous vehicle decision making. The application is wrapped within a user-friendly web interface developed using HTML/CSS and the Flask framework, offering two distinct views: one for the webcam feed and another for the dash-cam feed. Users can observe the object detection process in action through interactive and dynamically updating video frames. The interface provides a seamless experience, enhancing the understanding of the environmental analysis performed by the system. This project not only showcases the technical capability of deep learning in real-time object detection but also emphasizes its practical implementation in enhancing autonomous vehicle safety. The system’s design allows for scalability and future integration into autonomous vehicle platforms, contributing to the advancement of self-driving technology and ensuring safer journeys for passengers and pedestrians alike.

This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.