Search Results

Research is under way to achieve autonomous driving (AD) on urban roads by 2020. This study focused on a right-turn situation at an intersection difficult to navigate even for human drivers. Many accidents have been reported when turning right due to poor visibility of traffic in an oncoming lane. An AD system is being developed that can recognize and avoid a potential risk of accident where another vehicle at an intersection causes poor visibility. The aim is to automate identification of vehicle locations and locations with poor visibility as well as decisions as to whether to decelerate, accelerate, or stop. As solutions, an obstacle detection frame (ODF) was created to identify the locations of other vehicles and a visibility detection frame (VDF) was created to identify the locations of blind spots. Detection frames are defined in an internal intersection map in the AD system.

This paper presents a new method for calculating optical flow using data from a high frame-rate camera. We focused on a feature of image data captured with a high frame-rate camera in which objects do not move more than one pixel between successive frames. This approach eliminates repetitive processing for object identification among frames taken at different sampling times. High-speed processing hardware architecture was designed with sequential processing only, and the algorithm was implemented in a field programmable gate array. The resultant unit can calculate optical flow for a 640×120 pixel size image with a 480-Hz processing cycle and 0.5-μsec processing latency.