Simple road side detection and tracking on embedded PC

Abstract

Road tracking is highly required to drive vehicle autonomously. However, the tracking system must deal with two main constraints; real-time computation and limited computing power. Among various efforts, a road side detection is an alternative to provide road tracking. However, dealing road side detection with those constraints is still challenging. Therefore, this paper proposes a simple road side detection and tracking which is adequate for low cost embedded PC. Mainly, the road appearance has different texture to the road side. An algorithm is developed by searching the changing texture from the center to the road side. Here, this paper observes road existence about half part of road appearance on image. Which is about 10 ~ 14m ahead. It depends on perspective appearance of hardware setup. This interested region is observed horizontally on certain consecutive distance. Later, projection on image plane of each corresponding distance are observed for changing texture. The proposed method is evaluated within the first 500 consecutive image frames from the starting time on moving car. Four trials with different tracks are considered for evaluation. The algorithm was successfully run with 640x480 image resolution on raspberry Pi with accuracy 87%. The image was taken with 15fps and the computation for road detection was not more than 10ms.