At intersections, road features related to different maneuvers, such as left-turn, right-turn, and central channelization (i.e., guidelines and channelized islands), are widely used to decrease the traffic conflicts and improve the safety and mobility of traffic. However, there are several main problems related to channelization design at intersections, including poor recognizability, unreasonable entering speed, insufficient sight distance, and relatively low merging speed. To address these problems, this paper focused on developing a method to assess how to design intersection features by dividing the driving process through intersections into four stages: ''appearance of channelization,'' ''beginning of channelization,'' ''middle of channelization,'' and ''end of channelization.'' Drivers' visual lane models were established based on the Catmull-Rom spline to quantify visual road information perceived by drivers. Shape parameters and three characteristic regions were extracted from these models. The naturalistic driving experiments and 3D mobile mapping experiments were conducted at intersections with channelized islands or guidelines. Driving speed distributions in the four stages were found to obey normal distributions and could be calculated by shape parameters with Bayesian inference. Then, 3D mobile mapping was used as a substitute for extensive naturalistic driving experiments to obtain drivers' visual perception from all possible visual angles. An evaluation method on intersection design was built and used to identify which stage of channelization needs to be modified. This new method helps to enhance the safety and efficiency of intersections. INDEX TERMS 3D mobile mapping, intersection design, channelization, drivers' visual lane model, Bayesian inference.