Advanced Driver Assistance System for Occlusion Removal : From System Design to Software-in-the-Loop Testing

Read full text Summarise with VIRAT
Authors: Arttu Leppäaho
Cited by: 0
References: 30

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Cars have advanced far since their invention in the 19th century, but road safety remains a challenge even in the present day. Electronics are continuously playing a larger role in the pursuit of safer driving, a trend partially driven by the recent proliferation of autonomous vehicles. Sensors in particular are vital in enabling a vehicle to operate autonomously without human intervention, but they also enable new kinds of advanced driver assistance systems (ADAS) which—as the term implies—assist a human driver without assuming full control of the vehicle. This thesis explores the idea of expanding a driver’s field of vision by using digital image processing to make occluding objects in the environment transparent. This technique has the potential to prevent accidents by allowing the driver to react to things that would not otherwise be visible. The transparency effect can be achieved through depth-image-based rendering (DIBR), which requires image data from different viewpoints in the environment around the vehicle. The acquired data is combined into a 3D model that contains more spatial information of the environment than would fit into a flat 2D image. An intuitive way to visualize the 3D model is to show it from the driver’s perspective without drawing any of the occlusions, allowing the driver to see through them. Using the driver’s own perspective makes it easier to understand the true locations of the objects in the image, which is important so that the driver can parse the view quickly and react immediately if the situation demands it. As part of this thesis, a software solution is proposed for generating see-through views using DIBR. The software had to be lightweight enough to run in real time on a low-power embedded platform that could be reasonably placed inside a real car. An NVIDIA Jetson development board was used to represent such a platform. Furthermore, fast and reliable wireless communication was needed to share data with the vehicle. High Efficiency Video Coding (HEVC) was used to compress the sensor data and thus keep the network bandwidth manageable without sacrificing visual quality. Simulations cannot fully replace real-world testing, but they are a good way to preliminarily test vehicular systems without the risk of damage or injury should something go wrong. The proposed ADAS was tested in a software-in-the-loop (SIL) simulation with a simulation framework that replicated the real execution environment of the software by providing virtual sensors and traffic. The existing framework was extended with a flexible data processing module to achieve the desired functionality. SOLID software development principles were used to create an object-oriented design that is easy to extend for future use cases. Based on visual and performance evaluations, the proposed ADAS was found suitable for real-time use in low-velocity traffic assuming that sufficient sensor coverage and network availability can be provided. The latency of the system was measured to be around 32 ms when using 256×256 images. The see-through views are useful even at such low image resolutions, but the sensors must be positioned parallel to the viewing direction for the best results. All implemented software is cross-platform and exclusively uses open-source tools.

Fields of Study & Subjects

Autonomous Vehicle Technology and Safety Advanced Vision and Imaging Advanced Optical Imaging Technologies

Publication details

Journal / source
Tampere University Institutional Repository (Tampere University)
Publication date
1 January 2026
Language
EN
Document type
Thesis