Road Actor Recognition and Classification for First-Person Real-Time 3D Rendering

This thesis introduces a cutting-edge approach to enhance the perception capabilities of autonomous vehicles in real-world driving scenarios. The research aims to recognize and classify various road actors, including vehicles, pedestrians, cyclists, and obstacles, employing advanced computer vision techniques, and leveraging tools like Qt Quick 3D and C++ programming language. The study initiates with an in-depth review of existing methodologies in road actor recognition and classification, outlining their strengths, limitations, and avenues for improvement. Building upon this foundation, the thesis proposes an approach integrating state-of-the-art deep learning algorithms with real-time 3D rendering techniques, using Qt Quick 3D. The key part of this thesis is the creation and implementation of an algorithm capable of extracting environmental information using a stereo camera. This algorithm can recognize and classify all road actors within its field of view. The system leverages convolutional neural networks (CNNs) for feature extraction and classification, enabling it to understand the position and orientation of vehicles. Additionally, Qt Quick 3D is used to develop a human-machine interface (HMI) that provides real-time 3D representations of the road environment. The use of C++ ensures seamless integration and efficient execution of the proposed solution. The research outcome will contribute to the ongoing efforts towards developing safer and more efficient autonomous vehicles capable of navigating complex real-world environments with enhanced perception capabilities.