Lane detection is an operation in vision-based automotive safety applications. To reduce the computational complexity of lane detection, the region of interest (ROI) is a commonly used approach. However, the traditional rectangular ROI in the Hough space is prone to result in vibrated lane markings. To solve this problem, we present an elliptical ROI in the Hough space. The proposed ROI can not only increase the accuracy of lane-marking detection but also reduce the computational complexity of lane detection. In addition, we also present a new auxiliary lane detection algorithm to deal effectively with cases where one lane-marking is missing. Experimental results show that the proposed algorithm results in more accurate and satisfying lane-marking detection.