Robust detection and tracking of vehicle taillight signals using frequency domain feature based Adaboost learning

In this paper, we present a robust vision-based autonomous tracking of vehicle taillights and signal detection. In this study, the vehicle candidates are detected by Harr-like based classifier in bright scene and paired rear lights in dark scene, and this provides the taillight ROI for alert signal analysis. Then, the detected ROI regions are tracked with kernelized correlation filter. To recognize the taillight signal, we propose an invariant feature by finding the light scattering area with color and luminance analysis. The frequency response of dynamics and history of the light scattering area size can be trained by Adaboost classifier to detect the turn signal. Moreover, the high mounted stop light is used to detect the brake signal to improve the reliability of brake signal detection. The experimental results verified the accuracy and effectiveness of the proposed system.