Monitoring broiler feeding and drinking behaviors is essential for optimizing the health, productivity, and welfare of the poultry. However, it can be challenging due to the small size of broilers and the dynamic environment of poultry houses. This study proposes a broiler feeding and drinking behaviors monitoring system that uses improved YOLOv5 and DeepSORT to detect and track broilers, despite their relatively small size compared to other livestock. To better handle small objects in CCTV views, we leverage a query mechanism to accelerate the inference of YOLOv5. This mechanism initially predicts the coarse locations of small object on low-resolution features and then refines the detection results in high-resolution features sparsely guided by the coarse positions. In addition, we utilize shuffleNetv2 to accelerate the DeepSORT inference speed while maintaining accuracy. The proposed system was evaluated using videos footage from a commercial broiler farm, and the results showed that the system can detect and re-identification the broiler feeding and drinking behaviors with the mean average precision(mAP) of 97% when IOU threshold was set to 0.5. The proposed system has the potential to improve the welfare of broilers through enabling early intervention and broiler productivity by facilitating more effective management of feeding and drinking behaviors.