Beam selection is a crucial technology in wireless communication. Realizing highly mobile millimeter-wave and terahertz wireless communication necessitates the deployment of extensive massive antenna arrays within these systems and the utilization of narrow directional beams to mitigate path loss. However, adjusting the narrow beams of these antenna arrays can result in significant beam training overhead. To address this challenge, this paper proposes a machine learning-based vision-aided beam prediction method, grounded in an analysis of existing fundamental research methods. This approach employs wireless environmental images captured by base station cameras and leverages machine learning algorithms to train a model that predicts the optimal beam from a predefined codebook. A dataset comprising wireless environmental images from five different real-world scenarios, provided by the Wireless Intelligence Laboratory at ASU, was used for training the beam prediction model. The performance of the proposed model was compared with that of mainstream network models across various scenarios. Experimental results indicate that the proposed beam prediction model outperforms the baseline provided by the Wireless Intelligence Laboratory, validating its efficacy in enabling highly mobile millimeter-wave and terahertz wireless communication.