Vehicle Tracking Using Shape-Dependent Mixture Model With Edge-Concentrated Measurements

For tracking a rectangular vehicle, real-world automotive radar position measurements are distributed not uniformly over the vehicle extension but typically around the edges of the vehicle, i.e., the distribution of measurements is shape-dependent. To describe this phenomenon, a shape-dependent Gaussian mixture measurement model is presented, with each mixture component being used to describe a sub-rectangle region by introducing a shape scaling factor. The shape scaling factor is also shape-dependent and can characterize the measurement spread across the corresponding edge. In this model, parameters and mixture structure are highly shape-dependent, and the rectangular shape prior information is also incorporated. Based on the proposed model, a variational Bayesian approach is derived, which recursively and efficiently estimates the kinematic, shape, shape scaling factors, and orientation states of a vehicle. Additionally, the Doppler velocity measurement can also be integrated into the variational Bayesian framework by introducing a latent variable. This approach can effectively and adaptively describe the complex measurement distribution. From the simulation and real experimental results, the proposed approach has a great improvement in the tracking performance, and the superior performance of the proposed model is more significant in estimating the centroid position compared with the state-of-the-art approaches.