Lidar based on the optical phased array (OPA) and frequency-modulated continuous wave (FMCW) technology stands out in automotive applications due to its all-solid-state design, high reliability, and remarkable resistance to interference. However, while FMCW coherent detection enhances the interference resistance capabilities, it concurrently results in a significant increase in depth computation, becoming a primary constraint for improving point cloud density in such perception systems. To address this challenge, this study introduces a lidar solution leveraging the flexible scanning characteristics of OPA. The proposed system categorizes target types within the scene based on RGB images. Subsequently, it performs scans with varying angular resolutions depending on the importance of the targets. Experimental results demonstrate that, compared to traditional scanning methods, the target-adaptive method based on semantic segmentation reduces the number of points to about one-quarter while maintaining the resolution of the primary target area. Conversely, with a similar number of points, the proposed approach increases the point cloud density of the primary target area by about four times.