Road vanishing point detection is an important part of blind zone monitoring in advanced driver assistance systems. In view of the problems of low accuracy and large computation of existing vanishing point detection methods, a road vanishing point detection algorithm based on in-vehicle video images is proposed. The algorithm detected the image feature points based on Harris corner point detection by optimizing the score function to reduce the amount of operations in the tracking stage. It tracked the motion feature points by the pyramid optical flow method and frame difference distance, and accurately obtained the position of each feature point on the end frame. After removing the outlying points from the feature points, the K-Means clustering algorithm that optimized the initial clustering center was used to obtain the road vanishing point of the in-vehicle video image. Finally, the algorithm was applied to various vehicle driving scenes for testing. It can accurately detect the road vanishing points in the in-vehicle video images within a short running time, which proves that the algorithm is robust, simple and easy to implement.