Formulation of a vital monocular vision algorithm from video frames using a hybrid cascaded approach for disparity map acc

Ku-Fairolnizam, Ku Siti Nurul Ain (2025) Formulation of a vital monocular vision algorithm from video frames using a hybrid cascaded approach for disparity map acc. Project Report. Universiti Teknikal Malaysia Melaka, Melaka, Malaysia. (Submitted)

Text (Full Text) Formulation of a vital monocular vision algorithm from video frames using a hybrid cascaded approach for disparity map acc.pdf - Submitted Version Download (2MB)

Abstract

This paper proposes a critical monocular vision algorithm that uses a hybrid cascaded technique to improve the accuracy of disparity maps. The algorithm's goal is to use a single camera to mimic human vision in terms of depth perception, object recognition, and scene comprehension. Utilising methods like robust lane boundary detection, bird's-eye view transformations, and vehicle localisation, it makes use of developments in computer vision and artificial intelligence to maximise performance in real-time applications such as robotics, augmented reality, and autonomous driving. The study investigates cutting-edge techniques to address issues including low-texture areas, occlusions, radiometric aberrations, and intricate scene configurations, such as feature extraction, semantic segmentation, and deep learning models like convolutional neural networks (CNNs). The system can produce precise depth maps, identify lane boundaries, and identify cars with high accuracy, according to experimental validations. The project's use of a strong methodology allows for improved visual data interpretation in dynamic and difficult situations, such as changing road geometries, motion, and illumination. The algorithm's dependability and effectiveness are confirmed by testing on datasets and real-time situations. The study admits its limitations in dealing with extreme situations including lengthy distances, sharp pitch angles, and complex lane arrangements, notwithstanding these achievements. In order to increase resilience and flexibility, recommendations for future research place a strong emphasis on integrating multisensor data fusion, adaptive algorithms, and sophisticated AI techniques. This study makes a substantial contribution to the field of monocular vision systems, opening the door for their incorporation into intelligent systems for automated settings that are safer, more effective, and more sustainable.

Actions (login required)

View Item