This paper presents an improved multi-camera people tracking pipeline with a focus on enhancing object detection to improve overall tracking performance. The system integrates object detection, single-camera tracking, re-identification, and multi-camera matching. We evaluate five versions of the YOLO model, from YOLOv8 to YOLOv12 in the same tracking pipeline to ensure a fair comparison. Part of Track 1 data from AIC2024 dataset is used for both training and inference purposes. For each frame, detected objects are tracked using a Kalman filter and matched via the Hungarian algorithm based on both spatial and appearance features. Inter-camera identity association is achieved through cluster-based matching, followed by refinement steps to improve consistency. Tracking performance is assessed using the MOTA metric. Experimental results demonstrate that stronger detection models significantly improve tracking accuracy, with YOLOv12 achieving the highest MOTA score.