Projects

1. Lane shape estimation for driver assistant system (DAS)
    • We present a novel model based lane tracking framework, which has its strength in a new condition and improvement of various advanced methods. After we apply an inverse perspective mapping (IPM) to achieve top-view images of the street scene,  we model the shape of the street lanes using the linear-parabolic. Furthermore, the linear-parabolic model is estimated by the robust Partitioned Particle filter which is shown to be well appropriate for this task. The measurement function that describes how much one particle (model hypothesis) fits the image, or indicates how many local image descriptors support this particle, is modeled by the novel multiple kernel density. Previous measurement functions in general only consider color and edge, but here we also consider the gradient orientation as an important information for estimation. The experiments show the robustness of our new measurement model compared to the previous model. 

2. Robust nonlinear system estimation with improved numerical stability
    • The unscented information filter (UIF) has been introduced recently for nonlinear system estimation and sensor fusion. In the UIF framework, a number of sigma points are sampled from the probability distribution of the prior state by the unscented transform and then propagated through the nonlinear dynamic function and measurement function. The new state is estimated from the propagated sigma points. In this way, the UIF can achieve higher estimation accuracies and faster convergence rates than the extended information filter (EIF), which uses a Taylor series to linearize the nonlinear function. We extends the framework of the UIF: first, a central difference information filter (CDIF) is derived by employing Stirling's interpolation to generate the sigma points. This leads to fewer predefined parameters and lower computational cost as compared to the original UIF. Second, we introduce the square-root forms of the CDIF and UIF to increase the numerical stability and guarantee positive semi-definiteness of the state covariances. The proposed algorithms are finally evaluated on two nonlinear problems: the classical space-vehicle tracking problem and the bearing-only tracking problem.