In this paper, a new curve-lane recognition algorithm is proposed. The algorithm uses edge point curvature voting to determine the region of interest based on near-vision straight-lane information. First, information is detected in the near-vision area regarding the straight lines to the left and right of the current lane. Near-vision lane-line extraction includes lane image filtering, as well as edge detection of the region of interest below the vanishing line. The vanishing point is positioned by determining the position of the edge point and distribution of the direction angle. In addition, the straight line is extracted based on the position of the vanishing point. The straight lines that are constructed for the current lane in this way are selected and used as supplementation, in combination with the lane model. Next, the road curvature range isometry is divided into multiple subdivision regions. The near-vision lane straight-line curvature parameters extending from each edge point in the region of interest are computed by combining the straight-line near-vision lane information with the curve lane model in the pixel coordinate system. Subsequently, voting and counting are carried out for the curvature regions of each edge point to which the corresponding curvature computing values belong. Finally, the counting maximum from the corresponding curvature regions of the straight lines located to the left and right of the current lane are searched for, and the curvature region is converted, to obtain the lane line corresponding to the curvature parameter values. Experimental results indicate that the proposed curve-lane recognition algorithm can effectively detect the curve lanes of different curvatures. The results also indicate that the proposed curve detection method is highly accurate, and the algorithm is very robust in different environments.