Raoul Zöllner

An adaptive multi-layer particle filter for tracking of traffic participants in a roundabout

This paper describes an extension of a standard particle filter for application on multiple object tracking using a single probabilistic density function, at which the method is based on image sensor values coming from a monocular camera. The idea of the new approach is that each cell of the tracking area is associated with a so called layer, by using a new introduced layer distribution. The automatic adaption of the layer values occurs by the weights of the particles which are located in the cells of the tracking area. Depending on the layers, which are associated with the single particle positions, the particles get different areal extents. So, the posterior density is approximated using particles which have different areal extents. The principle of multiple layers is also used for finding new appearing objects via a reinitialization step. On basis of such a multi-layer particle filter, the tracking quality and robustness can be increased compared to the conventional method, especially when the measurement quality is low. The good performance is shown using a real traffic scene in a roundabout.

Hybridization of agriculture vehicles using a ride-on lawn mower

This paper discusses new powertrain concepts for outdoor robotic systems, which have to work energy autarkic over several hours under workload. The main idea is to separate the powertrain components into different devices (combustion and electric engines) according to functional and workload criteria, as well as device power characteristics. For demonstrating and validating the concept an existing ride-on lawn mower based on a combustion engine was modified by integration of two electric engines towards a hybrid machine. The achieved results show significant improvements fuel consumption and mowing power.

A Washout and a Tilt Coordination Algorithm for a Hexapod Platform

In this paper the modeling and simulation of a six degree of freedom hexapod platform simulator is presented. The simulator is used for vehicle driving simulations. Washout algorithms are used for the control of the platform. Components of the washout algorithms are low pass filters and high pass filters, as well as a tilt coordination algorithm. A test with realistic input acceleration data of a vehicle maneuver is performed to verify the model.

On the need and implementation of an open interface for the automotive domain

This paper describes a concept of open interfaces for the automotive domain. In recent years, several systems have been developed for connecting a car to the outside world by industry and academia. The paper will illustrate the need for an open interface to access the car by any mobile device. Furthermore, it discusses the requirements for those interfaces, security issues and commercialization aspects of the proposed system. In order to allow students to work on such interfaces a conceptual platform including the hardware and the software parts has been developed.

Multi-object Tracking Based on a Multi-layer Particle Filter for Unclustered Spatially Extended Measurements

The paper treats a multi-object tracking approach based on a multi-layer particle filter which is able to deal with the class of unclustered spatially extended measurements. The particle filter uses a so called adaptive layer distribution spanned over the tracking space, which determines the particles’ extents. Since the particle extents are used for the calculation of the particle weights, the multi-modal posterior distribution representing the dynamic objects is approximated with locally different resolutions. Moreover, the layer distribution is used to detect new appearing objects through a reinitialization step. In order to extract an object list out of the particle density, an Expectation Maximization (EM) clustering is used. The basic algorithm is extended with an estimation of the currently necessary number of clusters. The developed tracking approach is evaluated by means of image measurement data out of a roundabout scene. The proposed approach enhances tracking quality and robustness compared to a conventional approach.

Image-based multi-target tracking using a multi-layer particle filter and extended EM clustering

The paper presents an approach for tracking a variable number of objects by using a multi-layer particle filter combined with an extended Expectation Maximization (EM) clustering. The approach works on basis of binary foreground images coming from previous background subtraction. The multi-layer particle filter is an improvement of a conventional particle filter approach. It uses an introduced adaptive layer distribution spanned over the tracking area, which determines the areal extents of the particles. Thus, the multi-modal posterior distribution representing the objects is approximated with locally different resolutions. In addition, the layer distribution is used to find new appearing objects. In order to generate an object list out of the particle density, an EM clustering is used. The basic algorithm is extended with an estimation of the needful number of clusters by iteratively splitting and comparing the overall cluster areas. The new tracking approach improves tracking quality and robustness compared to the conventional particle filter approach. Experimental results are shown using the example of a traffic scene in a roundabout.

A modular and scalable simulation framework for agricultural vehicles

In the automotive area we have in our days a number of simulation environments that reproduce the dynamical characteristics of a vehicle in a very realistic way. These environments cannot be easily used in other domains or applied on complex, coupled machines for example ride-lawn mowers, for several reasons. On one hand the heterogeneous vegetation that has to be processed by the engine is influencing the dynamic driving characteristics of the vehicle and on the other hand the off-road terrain with inclinations are not normed and make the modelling process more difficult. This paper presents a concept of a simulation environment, where the agricultural environment is modelled as well as the driving dynamic characteristics of machines with coupled engines. The framework has real time capabilities and so enables Hardware in the Loop prototyping. We show this by validating a mowing process, where we explicitly model a hybrid ride-on lawn mower and the mowing environment.

A Method for Transforming Electric Vehicles to become Autonomous Vehicles

In order to develop autonomous navigation tasks for transportation systems or for fulfilling certain agriculture or industrial tasks, vehicle prototypes are needed. These vehicles cannot be reinvented every time. Existing vehicles are transformed by using a data processor that replaces the driver. Furthermore, depending on the vehicle type, actuators (e.g. steering or braking), sensors that observe the environment and interfaces to the information provided by the vehicle must be integrated. This process differs from vehicle to vehicle. In this paper we present a (further) solution for such a reconstruction, which is, indeed not completely general, but due to its modular architecture, the open software implementation and the use of open source components, easy to apply on new systems. The paper presents the hardware and software implementation on two vehicles. A third and a fourth one are planned. The electric power unit is not mandatory but helpful for the transformation, because an intervention in an already electric vehicle is easier to carry out.