<p>In this study, a piano keyboard detection approach was investigated. Because accurate recognition of a piano keyboard is linked to the accuracy of music transcription, piano keyboard detection is a fundamental step in research on music transcription and is a critical problem. Previous studies on piano keyboard detection have mainly focused on videos taken within a limited environment or taken above the piano. However, actual piano playing videos are often captured not only in limited environments, but also in various performance venues. Furthermore, in addition to filming above the piano, the piano is often filmed from the side. Using such videos capturing actual piano playing, it is difficult to accurately detect the piano keyboard through conventional computer vision technology. To address this challenge, this study introduces the Image Combining Method using Partial Pixel Information (ICM_PixInfo), which combines the partial pixel data of multiple images to enhance the accuracy of piano keyboard detection. Additionally, this study proposes the PD_BiTh algorithm, which extracts piano key positions from combined images. The performance of the proposed PiKd was compared with that of conventional methods. The results show that PiKd achieves the highest recall, reaching approximately 89.33%, and outperforms the comparative methods by about 21.05%p to 41.25%p on average.</p>