Abstract
To satisfy the requirements of road condition assessment an embedded lightfield system, combining 2D texture and 3D geometry, is required. An experimental prototype is proposed. Various approaches for lightfield acquisition were considered. In the end multi-view stereo was identified as the most suitable. Four camera pairs provide the raw data which is preprocessed and transmitted to the main processing unit, where the data is fused into one coherent dataset, via HDMI/DVI. Hybrid Zynq SoC technology was chosen to be able to support sequential and parallel computation paradigms. The system is adaptable to different fields of application due to its modular setup.
Highlights
IntroductionOne recent field is the road condition assessment
Automation and quality assurance reach into more and more fields of application
Early tests with a single camera pair showed promising results for asphalt, it is possible to extend the system with pseudorandom pattern projectors and additional cameras with spectral filters, should the textural contrast of the roads not suffice for reliable matching point detection and stereovision
Summary
One recent field is the road condition assessment. It is performed by a human expert evaluating image recordings of the road segments. Since this is not very economic and sometimes tedious and unattractive, there is incentive to aid the experts with computer vision technology. Various different approaches to lightfield acquisition were considered: structure from motion, (multi-view) stereo and plenoptic cameras. In structure from motion different perspectives of the same scene are provided by the same camera. It inherently has worse accuracy compared to a true stereo setting since the camera trajectory has to be estimated in addition to the scene geometry.
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
