<![CDATA[Conventional object detection systems in Intelligent Transportation Systems (ITS) commonly operate in a frame-wise manner, leading to temporally inconsistent predictions under dynamic conditions. This paper proposes a motion-consistent object detection framework that integrates velocity-constrained filtering to enforce physical plausibility across consecutive frames. The main contribution of this work lies in explicitly formulating detection validation as a deterministic velocity-bounded constraint problem, enabling direct integration of motion dynamics into the perception pipeline rather than treating motion as auxiliary information. The system employs a lightweight YOLO11s-based detector for visual perception and GPS-based motion estimation to provide real-world velocity constraints. Experiments are conducted on a traffic infrastructure dataset comprising 52,135 images across 33 classes and evaluated on an Intel NUC 13 Pro edge computing platform. Results demonstrate that the proposed method improves temporal stability compared to conventional frame-wise detection while maintaining strong detection performance, achieving an mAP@0.5 of 0.945 at 20 FPS. The findings indicate that incorporating explicit motion constraints enhances detection reliability without introducing significant computational overhead. However, performance may degrade under low-speed GPS noise and challenging visual conditions such as occlusion, highlighting limitations for future improvement.]]>
Copyrights © 2026
