Phd Forum: Real-Time Lane-Vehicle Detection for Advanced Driver Assistance on Mobile Devices

摘要

Advanced driver assistance system (ADAS) has attracted a lot of attention due to the fast growing industry of smart cars. While extensive literature exists on this topic, none of them considers the important fact that many vehicles today do not have powerful embedded electronics or cameras. To address this issue, we demonstrate a new framework that utilizes microprocessors in mobile devices with embedded cameras for advanced driver assistance. The main challenge that comes with this low cost solution is the dilemma between limited computing power and tight latency requirement, and uncalibrated camera and high accuracy requirement. Accordingly, we propose an efficient, accurate, flexible yet light-weight real-time lane and vehicle detection method and implement it on Android devices. Real road test results suggest that an average latency of 15 fps can be achieved with a high accuracy of 12.58 average pixel offset for each lane in all scenarios and 97% precision for vehicle detection.