Dynamical Tracking Of Surrounding Objects For Road Vehicles Using Linearly-arrayed Ultrasonic Sensors

Ultrasonic sensors have a potential of large-scale application in field of road environment perception due to low cost,wide detection angle,small near field blind zone and robustness to variant lighting conditions.Long distance ultrasonic sensors based tracking techniques would face two new challenges: anti interference ability declines and detection error increases because of a complexier environment;detection frequency for each sensor decreases and volumn of sensor readings decreases.To accomplish the task of effective object tracking using long distance ultrasonic sensors,a static sensor model was created based on experiment results.Tracking algorithms suitable for nonlinear measurements and applicable in dynamic object tracking were developed,and validated by field experiments.Firstly,an experiment focused on perception characteristics of a single ultrasonic sensor was conducted.Based on the data collected from the experiment,an static ultrasonic sensor model was built including a detection scope model,a model of chance of detection and a detection error model.A model of two quadratic curves was adopted for the detection scope model,the fitting accuracy was 0.2m.A detection scope varied by different shapes and materials was generated by fitting that of objects with different shapes and materials.A linearly decreasing model was adopted describing the rapidly decreasing of detection chance when the object approached the detection boader.An detection error model influenced by object shape,material and location,including distance and orientation was built.For object tracking in automotive applications,a layout of linearly-arrayed ultrasonic sensors with equal space was designed along with two firing sequences: mutual and serial.Two Kalman filter based tracking algorithms were developed for object tracking under two typical tracking scenarios,which were extended Kalman filter and Unscented Kalman filter.Finally the algorithms were validated by simulations and field tests.The simulation results showed: in mutual firing sequence,EKF and UKF can both realize good tracking performance under scenario 1 and 2,which was better than the advanced triangle method.No significant difference was found between EKF and UKF.In serial firing sequence,EKF and UKF were validated under scenario 1.A good tracking ability can be found which was much better than that of the advanced triangle method.Also the UKF was found to have better tracking performance than that of EKF.The field tests showed: in scenarios 1,EKF and UKF had good tracking ability which was better than that of the advanced triangle method.And the UKF was found to have better tracking performance than that of EKF.In this paper,a static sensor model was created based on experiment results.Tracking algorithms suitable for nonlinear measurements and applicable in dynamic object tracking were developed.Simulation results and field tests showed that the developed EKF and UKF had better tracking ability than the advanced triangle method.