Predicting the motion of traffic participants accurately remains a challenging task in the field of automated driving. Especially interactions between traffic participants introduce high complexity and interdependencies into the environment prediction. This work presents the remarkable performance of a Partially Observable Markov Decision Process (POMDP) framework to stochastically predict and safely respond to an interacting environment. The framework is validated for its ability to increase the overall Ego-Vehicle safety by preemptively triggering a de-escalation maneuver. The performance of the framework is analyzed on a publicly available dataset with real-world traffic (Argoverse) and on highly critical simulation scenarios specified by Euro-NCap for emergency braking functions. The results show quantitatively that the proposed framework significantly contributes to an early de-escalation of critical scenarios. Such an early de-escalation increases the safety and comfort of automated vehicles.

Trajectory prediction is crucial to advance autonomous driving, improving safety, and efficiency. Although end-to-end models based on deep learning have great potential, they often do not consider vehicle dynamic limitations, leading to unrealistic predictions. To address this problem, this work introduces a novel hybrid model that combines deep learning with a kinematic motion model. It is able to predict object attributes such as acceleration and yaw rate and generate trajectories based on them. A key contribution is the incorporation of expert knowledge into the learning objective of the deep learning model. This results in the constraint of the available action space, thus enabling the prediction of physically feasible object attributes and trajectories, thereby increasing safety and robustness. The proposed hybrid model facilitates enhanced interpretability, thereby reinforcing the trustworthiness of deep learning methods and promoting the development of safe planning solutions. Experiments conducted on the publicly available real-world Argoverse dataset demonstrate realistic driving behaviour, with benchmark comparisons and ablation studies showing promising results.

Precise vehicle state estimation is crucial for safe and reliable autonomous driving. The number of measurable states and their precision offered by the onboard vehicle sensor system are often constrained by cost. For instance, measuring critical quantities such as the Vehicle Sideslip Angle (VSA) poses significant commercial challenges using current optical sensors. This paper addresses these limitations by focusing on the development of high-performance virtual sensors to enhance vehicle state estimation for active safety. The proposed Uncertainty-Aware Hybrid Learning (UAHL) architecture integrates a machine learning model with vehicle motion models to estimate VSA directly from onboard sensor data. A key aspect of the UAHL architecture is its focus on uncertainty quantification for individual model estimates and hybrid fusion. These mechanisms enable the dynamic weighting of uncertainty-aware predictions from machine learning and vehicle motion models to produce accurate and reliable hybrid VSA estimates. This work also presents a novel dataset named Real-world Vehicle State Estimation Dataset (ReV-StED), comprising synchronized measurements from advanced vehicle dynamic sensors. The experimental results demonstrate the superior performance of the proposed method for VSA estimation, highlighting UAHL as a promising architecture for advancing virtual sensors and enhancing active safety in autonomous vehicles.

The ability to operate safely in increasingly complex traffic scenarios is a fundamental requirement for Automated Driving Systems (ADS). Ensuring the safe release of ADS functions necessitates a precise understanding of the occurring traffic scenarios. To support this objective, this work introduces a pipeline for traffic scenario clustering and the analysis of scenario category completeness. The Clustering Vector Quantized - Variational Autoencoder (CVQ-VAE) is employed for the clustering of highway traffic scenarios and utilized to create various catalogs with differing numbers of traffic scenario categories. Subsequently, the impact of the number of categories on the completeness considerations of the traffic scenario categories is analyzed. The results show an outperforming clustering performance compared to previous work. The trade-off between cluster quality and the amount of required data to maintain completeness is discussed based on the publicly available highD dataset.

Recent cooperative perception datasets have played a crucial role in advancing smart mobility applications by enabling information exchange between intelligent agents, helping to overcome challenges such as occlusions and improving overall scene understanding. While some existing real-world datasets incorporate both vehicle-to-vehicle and vehicle-to-infrastructure interactions, they are typically limited to a single intersection or a single vehicle. A comprehensive perception dataset featuring multiple connected vehicles and infrastructure sensors across several intersections remains unavailable, limiting the benchmarking of algorithms in diverse traffic environments. Consequently, overfitting can occur, and models may demonstrate misleadingly high performance due to similar intersection layouts and traffic participant behavior. To address this gap, we introduce UrbanIng-V2X, the first large-scale, multi-modal dataset supporting cooperative perception involving vehicles and infrastructure sensors deployed across three urban intersections in Ingolstadt, Germany. UrbanIng-V2X consists of 34 temporally aligned and spatially calibrated sensor sequences, each lasting 20 seconds. All sequences contain recordings from one of three intersections, involving two vehicles and up to three infrastructure-mounted sensor poles operating in coordinated scenarios. In total, UrbanIng-V2X provides data from 12 vehicle-mounted RGB cameras, 2 vehicle LiDARs, 17 infrastructure thermal cameras, and 12 infrastructure LiDARs. All sequences are annotated at a frequency of 10 Hz with 3D bounding boxes spanning 13 object classes, resulting in approximately 712k annotated instances across the dataset. We provide comprehensive evaluations using state-of-the-art cooperative perception methods and publicly release the codebase, dataset, HD map, and a digital twin of the complete data collection environment via https://github.com/thi-ad/UrbanIng-V2X.

In order to reliably validate autonomous driving functions, known risks must be taken into account and unknown risks must be identified. This work addresses this challenge by investigating risks at the level of object state estimations. The proposed methodology utilizes the differences between object state estimations from independent sensors, enabling the detection of relevant differences. This is a significant advantage, because sensor errors can be detected without ground truth. A deep autoencoder architecture is introduced to map the differences between state estimations into a latent space. The autoencoder contains Transformer and LSTM components to effectively process signals of varying lengths. The latent space is shaped using a k-means friendly design procedure, in order to find a suitable representation for anomaly detection. Detecting anomalies is a key component in the validation process of autonomous vehicles, contributing to the identification of unknown risks. The proposed approach is evaluated using real-world automotive sensor data from cameras and laser scanners in the publicly available nuScenes dataset. The results show that the generated latent space using the k-means friendly procedure is well suited for clustering differences between state estimations from these two sensors and thus for anomaly detection. In the framework specified in the safety standard ISO 21448 (SOTIF) the proposed methodology can play a key role for the detection of unknown risks on the perception level during the operation phase of autonomous vehicles.

This research focuses on solving challenges related to identifying unfamiliar object categories in the realm of Open-Set Object Detection (OSOD) using infrastructure sensors. Traditional camera-based OSOD systems struggle to generate proposals for dissimilar novel classes due to a lack of feature similarity. This research introduces a novel approach named Fusion Object Detector (FOD), which emphasizes the localization and identification of semantically dissimilar unknown objects through a multimodal fusion architecture involving infrastructure-mounted cameras and LiDARs. FOD leverages a camera-based closed-set object detector for the identification of known class objects, while simultaneously utilizing clusters derived from fused LiDAR point clouds for the detection of unknown class objects. This research work also presents a novel dataset named Thermal camera and LiDAR in Infrastructure Dataset (TLID). TLID comprises fused sensor measurements from multiple thermal cameras and LiDARs mounted in three urban crossings of Ingolstadt city and at CARISSMA outdoor test track. The proposed methodology is evaluated using both an in-house dataset and a publicly available infrastructure dataset for the task of OSOD. The results quantify the importance of multimodal sensor information for the task of identifying dissimilar unknown objects.

For automotive applications, the Graph Attention Network (GAT) is a prominently used architecture to include relational information of a traffic scenario during feature embedding. As shown in this work, however, one of the most popular GAT realizations, namely GATv2, has potential pitfalls that hinder an optimal parameter learning. Especially for small and sparse graph structures a proper optimization is problematic. To surpass limitations, this work proposes architectural modifications of GATv2. In controlled experiments, it is shown that the proposed model adaptions improve prediction performance in a node-level regression task and make it more robust to parameter initialization. This work aims for a better understanding of the attention mechanism and analyzes its interpretability of identifying causal importance.

Sensors installed in the infrastructure can make a significant contribution to the advancement of Advanced Driver Assistance Systems (ADAS) and connected mobility. Thermal cameras provide protection against the abuse of personalised data and perform robustly in challenging environmental conditions, making them an excellent choice for infrastructural perception. The goal of this work is to solve the crucial problem of Open-Set Recognition (OSR) for thermal camera-based perception systems installed in the infrastructure. In this paper, a novel modular architecture for OSR called Class Specific Experts (CSE) is proposed, in which, class specialization is achieved using individual feature spaces. The proposed methodology can be easily embedded in an object detection setting and provides as a main advantage, the possibility of online incremental learning without catastrophic forgetting. This work also introduces a open-source classification dataset called Infrastructure Thermal Dataset (ITD) containing image snippets captured by a thermal camera mounted in the infrastructure. The proposed approach outperforms the compared baselines for the task of OSR on many publicly available thermal and non-thermal datasets, as well as the new ITD dataset.

Clustering traffic scenarios and detecting novel scenario types are required for scenario-based testing of autonomous vehicles. These tasks benefit from either good similarity measures or good representations for the traffic scenarios. In this work, an expert-knowledge aided representation learning for traffic scenarios is presented. The latent space so formed is used for successful clustering and novel scenario type detection. Expert-knowledge is used to define objectives that the latent representations of traffic scenarios shall fulfill. It is presented, how the network architecture and loss is designed from these objectives, thereby incorporating expert-knowledge. An automatic mining strategy for traffic scenarios is presented, such that no manual labeling is required. Results show the performance advantage compared to baseline methods. Additionally, extensive analysis of the latent space is performed.

An understanding and classification of driving scenarios are important for testing and development of autonomous driving functionalities. Machine learning models are useful for scenario classification but most of them assume that data received during the testing are from one of the classes used in the training. This assumption is not true always because of the open environment where vehicles operate. This is addressed by a new machine learning paradigm called open-set recognition. Open-set recognition is the problem of assigning test samples to one of the classes used in training or to an unknown class. This work proposes a combination of Convolutional Neural Networks (CNN) and Random Forest (RF) for open set recognition of traffic scenarios. CNNs are used for the feature generation and the RF algorithm along with extreme value theory for the detection of known and unknown classes. The proposed solution is featured by exploring the vote patterns of trees in RF instead of just majority voting. By inheriting the ensemble nature of RF, the vote pattern of all trees combined with extreme value theory is shown to be well suited for detecting unknown classes. The proposed method has been tested on the highD and OpenTraffic datasets and has demonstrated superior performance in various aspects compared to existing solutions.

Traffic scenario categorisation is an essential component of automated driving, for e.g., in motion planning algorithms and their validation. Finding new relevant scenarios without handcrafted steps reduce the required resources for the development of autonomous driving dramatically. In this work, a method is proposed to address this challenge by introducing a clustering technique based on a novel data-adaptive similarity measure, called Random Forest Activation Pattern (RFAP) similarity. The RFAP similarity is generated using a tree encoding scheme in a Random Forest algorithm. The clustering method proposed in this work takes into account that there are labelled scenarios available and the information from the labelled scenarios can help to guide the clustering of unlabelled scenarios. It consists of three steps. First, a self-supervised Convolutional Neural Network (CNN) is trained on all available traffic scenarios using a defined self-supervised objective. Second, the CNN is fine-tuned for classification of the labelled scenarios. Third, using the labelled and unlabelled scenarios an iterative optimisation procedure is performed for clustering. In the third step at each epoch of the iterative optimisation, the CNN is used as a feature generator for an unsupervised Random Forest. The trained forest, in turn, provides the RFAP similarity to adapt iteratively the feature generation process implemented by the CNN. Extensive experiments and ablation studies have been done on the highD dataset. The proposed method shows superior performance compared to baseline clustering techniques.

Detecting unknown and untested scenarios is crucial for scenario-based testing. Scenario-based testing is considered to be a possible approach to validate autonomous vehicles. A traffic scenario consists of multiple components, with infrastructure being one of it. In this work, a method to detect novel traffic scenarios based on their infrastructure images is presented. An autoencoder triplet network provides latent representations for infrastructure images which are used for outlier detection. The triplet training of the network is based on the connectivity graphs of the infrastructure. By using the proposed architecture, expert-knowledge is used to shape the latent space such that it incorporates a pre-defined similarity in the neighborhood relationships of an autoencoder. An ablation study on the architecture is highlighting the importance of the triplet autoencoder combination. The best performing architecture is based on vision transformers, a convolution-free attention-based network. The presented method outperforms other state-of-the-art outlier detection approaches.

The availability of real-world data is a key element for novel developments in the fields of automotive and traffic research. Aerial imagery has the major advantage of recording multiple objects simultaneously and overcomes limitations such as occlusions. However, there are only few data sets available. This work describes a process to estimate a precise vehicle position from aerial imagery. A robust object detection is crucial for reliable results, hence the state-of-the-art deep neural network Mask-RCNN is applied for that purpose. Two training data sets are employed: The first one is optimized for detecting the test vehicle, while the second one consists of randomly selected images recorded on public roads. To reduce errors, several aspects are accounted for, such as the drone movement and the perspective projection from a photograph. The estimated position is comapared with a reference system installed in the test vehicle. It is shown, that a mean accuracy of 20 cm can be achieved with flight altitudes up to 100 m, Full-HD resolution and a frame-by-frame detection. A reliable position estimation is the basis for further data processing, such as obtaining additional vehicle state variables. The source code, training weights, labeled data and example videos are made publicly available. This supports researchers to create new traffic data sets with specific local conditions.

Due to their capability of acquiring aerial imagery, camera-equipped Unmanned Aerial Vehicles (UAVs) are very cost-effective tools for acquiring traffic information. However, not enough attention has been given to the validation of the accuracy of these systems. In this paper, an analysis of the most significant sources of error is done. This includes three key components. First, a vehicle state estimation by means of statistical filtering. Second, a quantification of the most significant sources of error. Third, a benchmark of the estimated state compared with state-of-the-art reference sensors. This work presents ways to minimize the errors of the most relevant sources. With these error reductions, camera-equipped UAVs are very attractive tools for traffic data acquisition. The test data and the source code are made publicly available.

A novel unsupervised outlier score, which can be embedded into graph based dimensionality reduction techniques, is presented in this work. The score uses the directed nearest neighbor graphs of those techniques. Hence, the same measure of similarity that is used to project the data into lower dimensions, is also utilized to determine the outlier score. The outlier score is realized through a weighted normalized entropy of the similarities. This score is applied to road infrastructure images. The aim is to identify newly observed infrastructures given a pre-collected base dataset. Detecting unknown scenarios is a key for accelerated validation of autonomous vehicles. The results show the high potential of the proposed technique. To validate the generalization capabilities of the outlier score, it is additionally applied to various real world datasets. The overall average performance in identifying outliers using the proposed methods is higher compared to state-of-the-art methods. In order to generate the infrastructure images, an openDRIVE parsing and plotting tool for Matlab is developed as part of this work. This tool and the implementation of the entropy based outlier score in combination with Uniform Manifold Approximation and Projection are made publicly available.

Autonomous driving is an important trend of the automotive industry. The continuous research towards this goal requires a precise reference vehicle state estimation under all circumstances in order to develop and test autonomous vehicle functions. However, even when lane-accurate positioning is expected from oncoming technologies, like the L5 GPS band, the question of accurate positioning in roofed areas, e.g., tunnels or park houses, still has to be addressed. In this paper, a novel procedure for a reference vehicle state estimation is presented. The procedure includes three main components. First, a robust standstill detection based purely on signals from an Inertial Measurement Unit. Second, a vehicle state estimation by means of statistical filtering. Third, a high accuracy LiDAR-based positioning method that delivers velocity, position and orientation correction data with a mean error of 0.1 m/s, 4.7 cm and 1° respectively. Runtime tests on a CPU indicates the possibility of real-time implementation.

The goal of this paper is to provide a method, which is able to find categories of traffic scenarios automatically. The architecture consists of three main components: A microscopic traffic simulation, a clustering technique and a classification technique for the operational phase. The developed simulation tool models each vehicle separately, while maintaining the dependencies between each other. The clustering approach consists of a modified unsupervised Random Forest algorithm to find a data adaptive similarity measure between all scenarios. As part of this, the path proximity, a novel technique to determine a similarity based on the Random Forest algorithm is presented. In the second part of the clustering, the similarities are used to define a set of clusters. In the third part, a Random Forest classifier is trained using the defined clusters for the operational phase. A thresholding technique is described to ensure a certain confidence level for the class assignment. The method is applied for highway scenarios. The results show that the proposed method is an excellent approach to automatically categorize traffic scenarios, which is particularly relevant for testing autonomous vehicle functionality.

Due to the current developments towards autonomous driving and vehicle active safety, there is an increasing necessity for algorithms that are able to perform complex criticality predictions in real-time. Being able to process multi-object traffic scenarios aids the implementation of a variety of automotive applications such as driver assistance systems for collision prevention and mitigation as well as fall-back systems for autonomous vehicles. We present a fully model-based algorithm with a parallelizable architecture. The proposed algorithm can evaluate the criticality of complex, multi-modal (vehicles and pedestrians) traffic scenarios by simulating millions of trajectory combinations and detecting collisions between objects. The algorithm is able to estimate upcoming criticality at very early stages, demonstrating its potential for vehicle safety-systems and autonomous driving applications. An implementation on an embedded system in a test vehicle proves in a prototypical manner the compatibility of the algorithm with the hardware possibilities of modern cars. For a complex traffic scenario with 11 dynamic objects, more than 86 million pose combinations are evaluated in 21 ms on the GPU of a Drive PX 2.

In this paper, a probabilistic space-time representation of complex traffic scenarios is predicted using machine learning algorithms. Such a representation is significant for all active vehicle safety applications especially when performing dynamic maneuvers in a complex traffic scenario. As a first step, a hierarchical situation classifier is used to distinguish the different types of traffic scenarios. This classifier is responsible for identifying the type of the road infrastructure and the safety-relevant traffic participants of the driving environment. With each class representing similar traffic scenarios, a set of Random Forests (RFs) is individually trained to predict the probabilistic space-time representation, which depicts the future behavior of traffic participants. This representation is termed as a Predicted-Occupancy Grid (POG). The input to the RFs is an Augmented Occupancy Grid (AOG). In order to increase the learning accuracy of the RFs and to perform better predictions, the AOG is reduced to low-dimensional features using a Stacked Denoising Autoencoder (SDA). The excellent performance of the proposed machine learning approach consisting of SDAs and RFs is demonstrated in simulations and in experiments with real vehicles. An application of POGs to estimate the criticality of traffic scenarios and to determine safe trajectories is also presented.

Many variants of the Rapidly-exploring Random Tree (RRT) algorithm use biased-sampling strategies for solving computationally intensive tasks. One of such tasks is the planning of safe trajectories with the simultaneous intervention in both the longitudinal and the lateral dynamics of the vehicle in complex traffic-scenarios with multiple static and dynamic objects. A recently proposed hybrid statistical learning approach uses a 3D convolutional neural network (3D-ConvNet) to predict suitable longitudinal acceleration profiles in combination with an RRT variant called the Augmented CL-RRT algorithm. This algorithm is not effective in complex traffic-scenarios, i.e., traffic scenarios with more than 4 dynamic objects, because of the lack of flexibility and biasing in the longitudinal and the lateral dynamics intervention, respectively. Therefore, an extension to the Augmented CL-RRT algorithm is introduced to improve the longitudinal dynamics intervention with actuator and stable profile constraints and named as the Augmented CL-RRT+ algorithm. A biased-sampling strategy is also proposed based on the predicted longitudinal acceleration and steering wheel angle profiles provided by a trained 3D-ConvNet. Simulations are performed to compare different trajectory planning algorithms based on efficiency and safety. The results show vast improvements in terms of the efficiency without harming the safety.

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.

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.

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.

Advanced proactive safety applications are considered a promising approach to increase the effectiveness of already highly optimized vehicular safety systems. Detecting an unavoidable crash situation before the actual collision is of utmost importance and requires an effective real-time implementation. In this paper a collision detection algorithm based on the curvilinear-motion model for trajectory estimation is presented. The algorithm takes into account the EGO-vehicle's driving state and the high-level representation of surrounding objects. Next the presented approach is evaluated from a real-time perspective by applying static code analysis to a reference implementation of the algorithm. The results suggest the application of further optimization techniques as the computational complexity does not allow an effective real-time behavior. In order to guarantee both real-time constraints and effective collision detection a novel method for the preselection of potential collision opponents based on the Random Forest classifier is employed. The combination of efficient preselection and the proposed collision detection algorithm leads to a highly effective context interpretation that does not neglect the tight economic constraints.

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.

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