Scalable pipelines for instant object detection: deploying YOLO models with Apache Kafka for high-throughput inference

Instant object detection is a critical capability in modern applications where timely decision-making is essential, such as in emergency medicine, autonomous systems, and intelligent surveillance. Efficiently handling high-throughput image streams with minimal delay presents significant challenges, particularly under demanding conditions. This study presents a scalable object detection pipeline that integrates YOLO models with Apache Kafka, a distributed streaming platform, to support just-in-time inference. The proposed architecture leverages Kafka’s partitioning and consumer group mechanisms to enable parallel processing, ensuring high throughput without requiring complex load-balancing logic. The system is deployed on a Virtual Private Server to demonstrate practical implementation. Two configurations are presented to illustrate Kafka’s native scalability: one with a single partition and a single consumer, and another with five partitions and five consumers. These setups visually demonstrate how Kafka efficiently distributes workloads across multiple consumers. Although specific latency or throughput metrics are not reported, the architecture effectively showcases how Kafka’s design enables prompt responses to high-volume input. This pipeline is well-suited for time-sensitive object detection tasks and can be extended to a wide range of instant analytics applications where rapid feedback is critical.

1. Redmon, J., Divvala, S., Girshick, R., & Farhadi, A. (2016). You Only Look Once: Unified, Real-Time Object Detection. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp. 779-788.

2. Han, S., Lee, H., & Park, Y. (2018). A Real-Time Object Detection System Based on Cloud and Edge Computing. In International Conference on Cloud Computing and Big Data (CloudCom), pp. 153-160.

3. Shi, Y., Ding, G., & Wu, Q. (2019). Edge Computing: Vision and Challenges. IEEE Internet of Things Journal, 6(3), 4724-4737.

4. Apache Kafka. Apache Kafka Documentation. Available at: https://kafka.apache.org/39/documentation.html (Accessed: May 2025).

5. J. Redmon and A. Farhadi, “YOLO9000: Better, Faster, Stronger,” 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). IEEE, pp. 6517-6525, Jul. 2017. doi: 10.1109/cvpr.2017.690.

6. A. Soni and A. Rai, “YOLO for Medical Object Detection (2018-2024),” 2024 IEEE 3rd International Conference on Electrical Power and Energy Systems (ICEPES). IEEE, pp. 1-7, Jun. 21, 2024. doi: 10.1109/icepes60647.2024.10653506.

7. C.M. Badgujar, A. Poulose, and H. Gan, “Agricultural object detection with You Only Look Once (YOLO) Algorithm: A bibliometric and systematic literature review,” Computers and Electronics in Agriculture, vol. 223, p. 109090, Aug. 2024, doi: 10.1016/j.compag.2024.109090.

8. H. Chen, “YOLO Algorithm in Analysis and Design of Athletes’ Actions in College Physical Education,” 2024 International Conference on Interactive Intelligent Systems and Techniques (IIST). IEEE, pp. 764-768, Mar. 04, 2024. doi: 10.1109/iist62526.2024.00002.

9. R. Girshick, “Fast R-CNN,” 2015 IEEE International Conference on Computer Vision (ICCV). IEEE, Dec. 2015. doi: 10.1109/iccv.2015.169.

10. Ultralytics, "YOLOv12 Models," Ultralytics Documentation. Available at: https://docs.ultralytics.com/tr/models/yolo12/. (Accessed: May 2025).

11. Artillery Docs. Available at: https://www.artillery.io/docs (Accessed: May 2025).