State estimation of moving objects based on ellipses

State estimation of moving objects in complex environments is the foundation for robot cognition and intelligence. This study focuses on targets identification and states estimation utilizing a single laser scanner, which adopts multiple ellipses to model any possible object in the field of view of the sensor. First, a nearest neighbor clustering algorithm is used to achieve the segments of the data frame. Each segment represents a potential target that consists of a group points close to each other in Euclidean distance. Then, an ellipse can be optimized with covering the group points by minimizing the ellipse area. The solution can be obtained by solving a traditional Lowner-John ellipsoid problem. The movement of the target will be replaced by the motion of the corresponding ellipse. Next, the target states are estimated by data association technique on their frame sequences. Finally, experiments with a still or moving robot mounted with a laser sensor are taken indoor and outdoor to verify the feasibility and effectiveness of our novel method.