Accurate real-time object counting on industrial conveyor belts is essential for automated quality control and production monitoring. Conventional computer vision methods, including background subtraction and template matching, exhibit limited robustness to environmental variations and occlusion scenarios prevalent in manufacturing environments. While tracking-based approaches such as DeepSORT and ByteTrack achieve higher accuracy, they introduce significant computational overhead unsuitable for resource-constrained industrial deployments. This paper presents an object counting system that integrates YOLOv11 detection with a novel zone-based counting algorithm designed for high-speed conveyor applications. Unlike single-line crossing methods that are sensitive to detection flickering, the proposed three-zone approach monitors spatial occupancy transitions, providing inherent robustness against momentary detection failures. The algorithm eliminates explicit object tracking while enabling bidirectional counting with O(n) computational complexity. A cooldown mechanism prevents double-counting errors during zone transitions. Experimental evaluation using Hikrobot industrial cameras demonstrated 94.2% counting accuracy at 28.5 frames per second on an NVIDIA RTX A3000 GPU. The system maintained 85.1% accuracy under 50–70% object occlusion and consistent performance across conveyor speeds ranging from 0.1 to 2.0 m/s. Comparative analysis against baseline methods, including traditional approaches and state-of-the-art trackers, confirmed statistically significant improvements (p < 0.001). Long-term reliability testing over 720 h validated stable operation suitable for industrial deployment. The results demonstrate that zone-based counting provides an effective balance between accuracy and computational efficiency, offering a practical solution for real-time industrial monitoring without requiring expensive tracking algorithms.