With the expansion of poultry farming scale, the automatic counting of chicken flocks has become a key requirement in modern farming models. Currently, vision-based counting methods have attracted increasing attention. However, many challenges remain in practical applications, such as occlusion, motion blur, and turning around of chicken flocks, which significantly hinder accurate counting. To address these issues, this paper proposes a dynamic counting method capable of handling large-scale chicken populations. Specifically, chicken counting is formulated as a Multi-Object Tracking (MOT) task based on the Tracking-by-Detection (TBD) framework. Firstly, a dedicated chicken dataset is constructed to train the detector, and the chickens in test videos are manually annotated for subsequent validation of the proposed method. Secondly, You Only Look Once version 11 Nano (yolov11n) is used as a detector to detect the position of chickens, and the improved Simple Online and Realtime Tracking (SORT) algorithm is used to track chickens. Finally, a double counting regions strategy is designed to solve the chicken turning problem and reduce the dependence of counting tasks on the model. The results show that the proposed method can achieve more than 98% counting accuracy, and supports real-time performance on resource-constrained devices.