Image fusion for robust 3D reconstruction of traffic scenes
|Coordinator||Volvo Personvagnar Aktiebolag - Avd 94000, PV32|
|Funding from Vinnova||SEK 1 530 000|
|Project duration||January 2014 - December 2015|
|Venture||Electronics, software and communication - FFI|
Purpose and goal
The aim of the project is to increase robustness of sensors used in active safety systems. By using advanced image processing algorithms, cameras are shown to be a good complement to radar and lidar systems thus complementing some of their weaknesses. It is demostrated that the methods can identify moving objects such as vehicles, pedestrians, animals etc.
Results and expected effects
The conducted research has demonstrated the benefit of using ordinary camera hardware in novel ways, extracting extra information from exposure control and spectral sensitivity of the image sensor in order to estimate the environment. Particular focus has been on developing methods for robust optical flow estimation based on image sequences with varying exposure settings and sensor methods providing diversity in the spectral domain. The results will be used in development of automotive active safety systems.
Approach and implementation
The research methodology is based on literature review, algorithm development based on statistical methods and tools from optimization theory. The derived algorithms are demonstrated on both synthetic image date as well as image sequences collected from vehicles in the real world.