Multiple Object Tracking (MOT) remains a critical challenge in complex scenarios characterized by non-linear motion and frequent occlusions. Traditional approaches like SORT rely on linear assumptions, leading to fragmentation and identity switches in dynamic scenes. This paper proposes SORT-MID, an enhanced framework that integrates the Motion Indeterminacy Diffusion (MID) model with adaptive mechanisms. The key innovations include a dynamic weight fusion method that balances predictions from the MID model and Kalman filter based on quantified prediction indeterminacy, and a velocity-aware trajectory adjustment mechanism that adaptively modulates historical input length to capture motion trends. By leveraging MID’s stochastic trajectory sampling, the framework effectively models non-linear dynamics while preserving much of the computational efficiency of Kalman filtering. Experimental results on the GMOT-40 dataset confirm the effectiveness of our method, which achieves competitive tracking performance in complex scenarios.