Towards a Transfer Concept from Camera-Based Driver Assistance to the Assistance of Visually Impaired Pedestrians

Abstract

In order to support visually impaired people in traffic situations, a concept for transferring existing camera-based object detection algorithms from Advanced Driver Assistance Systems (ADAS) to Assistive Systems for the Visually Impaired (ASVI) is in development. This paper summarizes the previous work towards a transfer concept, introduces research categories and according statements, and presents the evaluation of the three so far addressed use cases, namely road background segmentation, crosswalk, and lane detection. Based on a literature review, it is stated that ADAS and ASVI are similarly composed and that there is an overlap in use cases addressed in ADAS and needed in ASVI. The conduction and evaluation of a qualitative interview study makes it possible to determine the exact overlap: Lane, crosswalk, traffic sign, traffic light (state), (driving) vehicle, obstacle, and bicycle detection. Additionally, road background segmentation is considered. For every of these use cases, it has to be established if an adaptation is possible and leads to similar hit rates as the underlying algorithm. This is confirmed for the so far addressed use cases. The hit rates on the tested data are 99.87 % for road background segmentation, 98.64 % for crosswalk, and 97.89 % for lane detection. In the future, the remaining use cases will be examined and the findings will be summarized in a transfer concept.