In recent years, the escalating issue of non-renewable resource depletion has prompted a significant shift towards wind energy, recognized as a clean and renewable energy source, which has increasingly become a focal point for energy development. The accurate estimation of wind turbine power is typically contingent upon wind speed data; however, traditional methodologies rely on wind speed sensors that are not only costly but also often incapable of facilitating real-time monitoring under certain conditions at wind farm sites. To address this challenge, the present study proposes a video-based method for estimating wind turbine power, utilizing the YOLOv10-C2B framework in conjunction with an enhanced version of DeepSORT, thereby enabling real-time and precise power estimation. This research employs the YOLOv10-C2B model, which integrates the C3Ghost, BiFPN, and BiFormer modules to enhance small target performance. The experimental findings indicate that the improved model achieves a Precision of 96.7%, Recall of 76.9%, and mAP of 74.3%, while maintaining a model size of 5.2 MB, thus demonstrating significant performance optimization. To mitigate the ID jumping issue frequently encountered in target tracking, this study introduces the Track Buffer module, which further enhances the accuracy and stability of ID allocation, thereby minimizing the interference caused by misassignments. Ultimately, by conducting image processing on the ID-framed wind turbine blades to analyze angular velocity, the study achieves precise calculations of rotational angular velocity, facilitating real-time power estimation of wind turbines. The experimental results substantiate that this method can efficiently and accurately estimate the instantaneous power output of wind turbines, thereby providing valuable scientific.