Abstract: To enhance the visual perception of vehicles, an improved contour angle orientation (CAO) algorithm is proposed for the registration of infrared and visible light images in traffic scenes. By simulating different traffic scenarios, a performance comparison was conducted among mature algorithms to select the superior CAO algorithm. Subsequently, improvements were made to the coarse matching parameters and image preprocessing scaling procedure. Experiments demonstrate that the refined CAO algorithm achieves more precise fine matching, thus resulting in mosaic stitching with smoother transitions and lines and yielding better results. Compared with the original CAO algorithm, the improved version reduces the RMSE value by 3.29%, increases the precision value by 2.13%, and decreases the average computation time by 0.11 s, thereby demonstrating improvements in both registration accuracy and real-time performance.