Perception system for vehicle braking performance increase

Abstract

The context of this paper and research regards the completion of a camera-based, surroundings detection system. This system was built upon the premises of using a single camera, facing the front of a vehicle and serving as both a frontal object detection system and a road type and conditions classification system. The main principle on which this study was based was implementing Convolutional Neural Networks (CNN) for depth perception and image recognition. Several methods of customising on existing algorithmic architectures of CNNs as well as image implementation were compared based on other researchers and different image datasets. Alterations, customisations and improvements were made on the backbone algorithms of MonoDepth2 and GoogleNet (Inception v1) architectures for image classification and object detection. These results were then implemented into a single algorithm which calls both trained methods into a unified classifier designed for real- time classification and detection. Finally, the main scope of this project was that of creating such a system as an improved and more cost-efficient part of the earlier version of the Anti-lock Braking System (ABS) or alternatively, in service of any other Driverō¸°€s Assistance System.In the 50 pages of this report, the sections have taken into account the methodological approach to achieving the two tasks, the literature review regarding overall background research and experimentation as well as the analysis of each of the CNNs used for these task. Finally, the main results are analyzed in the conclusion section. This paper also contains a number of tables and figures regarding comparisons of neural networks, illustrations and result data.