A REAL-TIME AND RELIABLE BLIND SPOT DETECTION METHOD FOR INTELLIGENT DRIVER ASSISTANCE SYSTEMS

147 views

Abstract

Advanced driver assistance systems (ADASs) play an important role in camera-only active safety systems and intelligent autonomous vehicles. For these applications, real-time and reliable detection performance is required. This paper aims to optimize the automobile detection speed and false alarm decrement for blind spot detection systems. Accordingly, we first propose the cascade-AdaBoost classifier along with our own sample datasets and training algorithm. In addition, in order to improve the detection speed, a regions of interests (ROIs) technique is also used to avoid extracting regions such as sky or regions of interests inconsistent with perspective, which generate the potential number of false alarms. The proposed method respectively achieves a speedup of at least 1.90x and 2.24x false alarm decrement compared to the conventional approach for high resolution images (720 x 480), with detection rate of 99.40% and a minor false detection rate of 4.08%. This makes the proposed method possible to be applied to real-time intelligent autonomous vehicles.