Michael Botsch

Situation aspect modelling and classification using the Scenario Based Random Forest algorithm for convoy merging situations

Advanced Driver Assistance Systems (ADAS) require an understanding of complex traffic situations. Such complexity can be handled by decomposing traffic situations into analyzeable subsets called Situation Aspects (SA). Since lots of situation analyzing problems result in classification tasks, the Scenario Based Random Forest (SBRF) algorithm is introduced into the field of ADAS situation analysis research. This classification method is designed to handle feature sets that develop over time and classification results that can be judged only by using complete scenarios instead of single time snap-shots. Furthermore, it has the advantage of using the out of bag (oob) estimation technique in order to perform feature selection. The problem of detecting a convoy merging traffic situation in real traffic scenarios serves as example to show the process of situation aspect modelling, feature selection and classification using the above mentioned methodology. It is demonstrated how the challenge of labelling changing SA can be solved using an undefined transition class and how this effects classification results. Because unbalanced data sets often occur in ADAS situation analysis, results on over- and downsampling strategies are described as well.

Feature Selection for Change Detection in Multivariate Time-Series

In machine learning the preprocessing of the observations and the resulting features are one of the most important factors for the performance of the final system. In this paper a method to perform feature selection for change detection in multivariate time-series is presented. Feature selection aims to determine a small subset which is representative for the change detection task from a given set of features. We are dealing with time-series where the classification has to be done on time-stamp level, although the smallest independent entity is a scenario consisting of one or more time-series. Despite this difficulty we will show how feature selection based on the generalization ability of a classifier can be realized by defining a cost function on scenario level. For the classification step in the feature selection process a modified random forest (RF) algorithm - which we will call scenario based random forest (SBRF) - is used due to its intrinsic possibility to estimate the generalization error. The excellent performance of the proposed feature selection algorithm will be shown in a car crash detection application

A Machine Learning Approach for the Segmentation of Driving Maneuvers and its Application in Autonomous Parking

Abstract A classification system for the segmentation of driving maneuvers and its validation in autonomous parking using a small-scale vehicle are presented in this work. The classifiers are designed to detect points that are crucial for the path-planning task, thus enabling the implementation of efficient autonomous parking maneuvers. The training data set is generated by simulations using appropriate vehicle-dynamics models and the resulting classifiers are validated with the small-scale autonomous vehicle. To achieve both a high classification performance and a classification system that can be implemented on a microcontroller with limited computational resources, a two-stage design process is applied. In a first step an ensemble classifier, the Random Forest (RF) algorithm, is constructed and based on the RF-kernel a General Radial Basis Function (GRBF) classifier is generated. The GRBF-classifier is integrated into the small-scale autonomous vehicle leading to excellent performance in parallel-, cross- and oblique-parking maneuvers. The work shows that segmentation using classifies and open-loop control are an efficient approach in autonomous driving for the implementation of driving maneuvers.

Vehicle rear detection in images with Generalized Radial-Basis-Function classifiers

A classification system for vehicle rear detection in images is presented. The classification system consists of an expert system for preselecting relevant regions in an image and a subsequent machine learning classifier. Both utilize specifically designed features for the application. The latter classifier is implemented by a generalized radial basis function (GRBF) algorithm to assure the interpretability of the detection system. The GRBF is constructed based on the kernel of an ensemble method classifier, the random forest (RF) algorithm in order to achieve a low generalization error. The high accuracy of the proposed technique for vehicle rear detection is validated using data from real traffic situations.

Construction of interpretable Radial Basis Function classifiers based on the Random Forest kernel

In many practical applications besides a small generalization error also the interpretability of classification systems is of great importance. There is always a tradeoff among these two properties of classifiers. The similarity measure in the input space as defined by one of the most powerful classifiers, the Random Forest (RF) algorithm, is used in this paper as basis for the construction of Generalized Radial Basis Function (GRBF) classifiers. Hereby, interpretability and a low generalization error can be achieved. The main idea is to approximate the RF kernel by Gaussian functions in a GRBF network. This way the GRBF network can be constructed to approximate the conditional probability of each class given a query input. Since each center in the GRBF is used for the representation of the distribution of a single target class in a localized area of the classifiers input space, interpretability can be achieved by taking account for the membership of a query input to the different localized areas. Whereas in most algorithms the pruning technique is used only to improve the generalization property, here a method is proposed how pruning can be applied to additionally improve the interpretability. Another benefit that comes along with the resulting GRBF classifier is the possibility to detect outliers and to reject decisions that have a low confidence. Experimental results underline the advantages of the classification system.

Supervised Learning via Optimal Control Labeling for Criticality Classification in Vehicle Active Safety

A core component of vehicle active safety algo-rithms is the estimation of criticality, which is a measure of the threat or danger of a traffic situation. Based on the criticality esti-mate, an active safety system can significantly increase passenger safety by triggering collision avoidance or mitigation maneuvers like emergency braking or steering. Interpreting criticality as the intensity of an evasion maneuver, we formulate a MinMax optimal control problem which incorporates moving obstacles and clothoidal lane constraints. We show how the solution of this optimal control problem can be used as a criticality labeling function to generate reference data sets for collision scenes. In order to achieve fast execution speeds, we present a supervised classification approach to criticality estimation. Using the Random Forest classifier with feature selection, we show that the criticality of combined braking and steering maneuvers can be predicted with high precision.

Robust and reduced-rank matrix Wiener filter based on the conjugate gradient algorithm

In this paper, we consider the block conjugate gradient (BCG) algorithm as a robust and reduced-rank implementation of the matrix Wiener filter (MWF). Stopping the BCG iterations before the algorithm converges, reduces the computational complexity as well as the performance loss due to channel estimation errors which is especially a problem in low sample support scenarios. We investigate the inherent robustness of the BCG method by deriving the corresponding filter factor matrix which is a well-known tool in the theory of ill-posed problems to describe and analyze regularizing effects. Moreover, we present a robust BCG algorithm where we applied the regularization method of diagonal loading in order to increase both robustness against estimation errors and flexibility in choosing the number of BCG iterations without decreasing performance. Finally, we use quasi-optimal loading instead of common heuristic choices. Simulation results of a frequency-selective multi-user single-input multiple-output (MU-SIMO) system show the improved performance of the robust BCG filter compared to the conventional MWF despite of the enormously reduced computational complexity.

A Hybrid Machine Learning Approach for Planning Safe Trajectories in Complex Traffic-Scenarios

Planning of safe trajectories with interventions in both lateral and longitudinal dynamics of vehicles has huge potential for increasing the road traffic safety. Main challenges for the development of such algorithms are the consideration of vehicle nonholonomic constraints and the efficiency in terms of implementation, so that algorithms run in real time in a vehicle. The recently introduced Augmented CL-RRT algorithm is an approach that uses analytical models for trajectory planning based on the brute force evaluation of many longitudinal acceleration profiles to find collision-free trajectories. The algorithm considers nonholonomic constraints of the vehicle in complex road traffic scenarios with multiple static and dynamic objects, but it requires a lot of computation time. This work proposes a hybrid machine learning approach for predicting suitable acceleration profiles in critical traffic scenarios, so that only few acceleration profiles are used with the Augmented CL-RRT to find a safe trajectory while reducing the computation time. This is realized using a convolutional neural network variant, introduced as 3D-ConvNet, which learns spatiotemporal features from a sequence of predicted occupancy grids generated from predictions of other road traffic participants. These learned features together with hand-designed features of the EGO vehicle are used to predict acceleration profiles. Simulations are performed to compare the brute force approach with the proposed approach in terms of efficiency and safety. The results show vast improvement in terms of efficiency without harming safety. Additionally, an extension to the Augmented CL-RRT algorithm is introduced for finding a trajectory with low severity of injury, if a collision is already unavoidable.

Probability estimation for Predicted-Occupancy Grids in vehicle safety applications based on machine learning

This paper presents a method to predict the evolution of a complex traffic scenario with multiple objects. The current state of the scenario is assumed to be known from sensors and the prediction is taking into account various hypotheses about the behavior of traffic participants. This way, the uncertainties regarding the behavior of traffic participants can be modelled in detail. In the first part of this paper a model-based approach is presented to compute Predicted-Occupancy Grids (POG), which are introduced as a grid-based probabilistic representation of the future scenario hypotheses. However, due to the large number of possible trajectories for each traffic participant, the model-based approach comes with a very high computational load. Thus, a machine-learning approach is adopted for the computation of POGs. This work uses a novel grid-based representation of the current state of the traffic scenario and performs the mapping to POGs. This representation consists of augmented cells in an occupancy grid. The adopted machine-learning approach is based on the Random Forest algorithm. Simulations of traffic scenarios are performed to compare the machine-learning with the model-based approach. The results are promising and could enable the real-time computation of POGs for vehicle safety applications. With this detailed modelling of uncertainties, crucial components in vehicle safety systems like criticality estimation and trajectory planning can be improved.

Maneuver segmentation for autonomous parking based on ensemble learning

A classification system for the segmentation of parking maneuvers and its validation using a small-scale autonomous vehicle are presented in this work. The classifiers are designed to detect points that are crucial for the path-planning task, thus enabling the implementation of efficient autonomous parking maneuvers. The training data set is generated by simulations using appropriate vehicle-dynamics models and the resulting classifiers are validated with the small-scale autonomous vehicle. To achieve both a high classification performance and a classification system that can be implemented on a microcontroller with limited computational resources, a two-stage design process is applied. In a first step an ensemble classifier, the Random Forest (RF) algorithm, is constructed and based on the RF-kernel a General Radial Basis Function (GRBF) classifier is generated. The GRBF-classifier is integrated into the small-scale autonomous vehicle leading to an excellent performance in both parallel- and cross-parking maneuvers.