Localization using High definition (HD) maps is one of the key solutions for high precision autonomous driving on the urban environment. For accurate HD map-based localization, more features included in HD maps are required to be detected while driving in real-time. Although there were a few studies about detecting road features to match with HD map, mostly used only a simple intensity filter or dimensionality filter, which is not robust in varying environments. This paper proposes a novel traffic sign detection algorithm in the 3D point cloud using a 2D image for robust 3D object detection. Deep learning 2D object detection is the first step, and then by using LiDAR-camera calibration data and outlier removal algorithm, 3D point cloud data of the traffic signs are collected. The performance evaluation was done using Waymo Open Dataset and we compared the proposed method with other 3D object detection results in the challenges board. The proposed method is the robust solution for any feature recognition and 3D detection included in the HD map for precise localization.