Refine
Keywords
- Fahrerassistenzsystem (18)
- Fußgänger (10)
- Radfahrer (9)
- Fahrrad (6)
- Autonomes Fahrzeug (4)
- Prognose (2)
- Sensortechnik (2)
- Verkehrsverhalten (2)
- 3D body pose (1)
- 3D human trajectory forecast (1)
Document Type
- Conference Proceeding (20)
- Article (2)
Language
- English (22)
Has Fulltext
- no (22)
We present an approach for predicting continuous pedestrian trajectories over a time horizon of 2.5 s by means of polynomial least squares approximation and multilayer perceptron (MLP) artificial neural networks. The training data are gathered from 1075 real urban traffic scenes with uninstructed pedestrians including starting, stopping, walking and bending in. The polynomial approximation provides an extraction of the principal information of the underlying time series in the form of the polynomial coefficients. It is independent of sensor parameters such as cycle time and robust regarding noise. Approximation and prediction can be performed very efficiently. It only takes 35 ms on an Intel Core i7 CPU. Test results show
28% lower prediction errors for starting scenes and 32% for stopping scenes in comparison to applying a constant velocity movement model. Approaches based on MLP without polynomial input or Support Vector Regression (SVR) models as motion predictor are outperformed as well.
This paper provides a method to forecast pedestrian trajectories by means of polynomial least-squares approximation and multilayer perceptron artificial neural networks for traffic safety applications. The approach uses camera based head tracking as input data to predict a continuous trajectory for a 2.5 s future time horizon. Training and test is performed using 1075 recorded tracks of uninstructed pedestrians in common public traffic situations, including many challenging scenarios like starting, stopping and bending in. The neural network approach
has the ability to handle these scenes by learning a single implicit
movement model independent of a specific motion type. The polynomial approximation provides an extraction of the principal
information of the underlying time series in the form of the
polynomial coefficients, high independence of input data, e.g.,
sample rate, and additional noise resistance. Our test results
show 24% lower prediction errors for starting scenes and 29%
for stopping scenes in comparison to a constant velocity Kalman
filter. Approaches using MLP without polynomial input and the
usage of Support Vector
In this article, we investigate the use of 3D human poses for trajectory forecasting of vulnerable road users (VRUs), such as pedestrians and cyclists, in road traffic. The forecast is based on past movements of the respective VRU and an important aspect in driver assistance systems and autonomous driving, which both could increase VRU safety. The 3D poses represent the entire body posture of the VRUs and can therefore provide important indicators for trajectory forecasting. In particular, we investigate the influence of different joint combinations and input sequence lengths of past movements on the accuracy of trajectory forecasts for pedestrians and cyclists. In addition, we divide VRU movements into the motion types wait, start, move, and stop and evaluate the results separately for each of them. Comparing it to an existing, solely head based trajectory forecast, we show the advantages of using 3D poses. With an input sequence length of 1.0 s, the forecasting error is reduced by 17.9 % for starting, 8.18 % for moving, and 11.0 % for stopping cyclists. For pedestrians, the error is reduced by 6.93 %, 2.73 %, and 5.02 %, respectively. With shorter input sequences, the improvements over the solely head based method remain for cyclists and even increase for pedestrians.
In future traffic scenarios, vehicles and other traffic participants will be interconnected and equipped with various types of sensors, allowing for cooperation based on data or information exchange. This article presents an approach to cooperative tracking of cyclists using smart devices and infrastructure-based sensors. A smart device is carried by the cyclists and an intersection is equipped with a wide angle stereo camera system. Two tracking models are presented and compared. The first model is based on the stereo camera system detections only, whereas the second model cooperatively combines the camera based detections with velocity and yaw rate data provided by the smart device. Our aim is to overcome limitations of tracking approaches based on single data sources. We show in numerical evaluations on scenes where cyclists are starting or turning right that the cooperation leads to an improvement in both the ability to keep track of a cyclist and the accuracy of the track particularly when it comes to occlusions in the visual system. We, therefore, contribute to the safety of vulnerable road users in future traffic.
In the future, vehicles and other traffic participants will be interconnected and equipped with various types of sensors, allowing for cooperation on different levels, such as situation prediction or intention detection. In this paper, we present a cooperative approach for starting movement detection of cyclists using a boosted stacking ensemble approach realizing feature- and decision-level cooperation. We introduce a novel method based on a three-dimensional convolutional neural network (CNN) to detect starting motions on image sequences by learning spatio-temporal features. The CNN is complemented by a smart device based starting movement detection originating from smart devices carried by the cyclist. Both model outputs are combined in a stacking ensemble approach using an extreme gradient boosting classifier resulting in a fast and yet robust cooperative starting movement detector. We evaluate our cooperative approach on real-world data originating from experiments with 49 test subjects consisting of 84 starting motions.
Early Start Intention Detection of Cyclists Using Motion History Images and a Deep Residual Network
(2018)
In this article, we present a novel approach to detect starting motions of cyclists in real world traffic scenarios based on Motion History Images (MHIs). The method uses a deep Convolutional Neural Network (CNN) with a residual network architecture (ResNet), which is commonly used in image classification and detection tasks. By combining MHIs with a ResNet classifier and performing a frame by frame classification of the MHIs, we are able to detect starting motions in image sequences. The detection is performed using a wide angle stereo camera system at an urban intersection. We compare our algorithm to an existing method to detect movement transitions of pedestrians that uses MHIs in combination with a Histograms of Oriented Gradients (HOG) like descriptor and a Support Vector Machine (SVM), which we adapted to cyclists. To train and evaluate the methods a dataset containing MHIs of 394 cyclist starting motions was created. The results show that both methods can be used to detect starting motions of cyclists. Using the SVM approach, we were able to safely detect starting motions 0.506 s on average after the bicycle starts moving with an F 1 -score of 97.7%. The ResNet approach achieved an F 1- score of 100% at an average detection time of 0.144 s. The ResNet approach outperformed the SVM approach in both robustness against false positive detections and detection time.
In this article, we present an approach to forecast trajectories of vulnerable road users (VRUs) including a numerical quantification of the uncertainty of the forecast. The uncertainty estimates are modeled as normal distributions by means of neural networks. Additionally, we present a method to evaluate the reliability of the forecasted uncertainty estimates, where we utilize quantile-quantile (Q-Q) plots, a graphical method to compare two distributions widely used in statistics. The positional accuracy is evaluated using Euclidean distances, in specific we use the average Euclidean error (AEE) and the average specific AEE (ASAEE). The model is trained and tested using a large dataset of 1311 cyclist trajectories, recorded at an urban intersection in real world traffic. Using this method, we achieve a similar positional accuracy compared to our previous work, where only positions are forecasted. The method is able to produce reliable uncertainty estimates for the motion types start, stop, turn left, and turn rightand produces underconfident uncertainty estimates for the motion types waitand move straight. Since uncertainties are not underestimated, the method can be used as a basis for trajectory planing in automated vehicles.
Vulnerable road users (VRUs, i.e. cyclists and pedestrians) will play an important role in future traffic. To avoid accidents and achieve a highly efficient traffic flow, it is important to detect VRUs and to predict their intentions. In this article a holistic approach for detecting intentions of VRUs by cooperative methods is presented. The intention detection consists of basic movement primitive prediction, e.g. standing, moving, turning, and a forecast of the future trajectory. Vehicles equipped with sensors, data processing systems and communication abilities, referred to as intelligent vehicles, acquire and maintain a local model of their surrounding traffic environment, e.g. crossing cyclists.
Heterogeneous, open sets of agents (cooperating and interacting vehicles, infrastructure, e.g. cameras and laser scanners, and VRUs equipped with smart devices and body-worn sensors) exchange information forming a multi-modal sensor system with the goal to reliably and robustly detect VRUs and their intentions under consideration of real time requirements and uncertainties. The resulting model allows to extend the perceptual horizon of the individual agent beyond their own sensory capabilities, enabling a longer forecast horizon. Concealments, implausibilities and inconsistencies are resolved by the collective intelligence of cooperating agents.
Novel techniques of signal processing and modelling in combination with analytical and learning based approaches of pattern and activity recognition are used for detection, as well as intention prediction of VRUs. Cooperation, by means of probabilistic sensor and knowledge fusion, takes place on the level of perception and intention recognition. Based on the requirements of the cooperative approach for the communication a new strategy for an ad hoc network is proposed. The evaluation is done using real data gathered with a research vehicle, a research intersection with public traffic and mobile devices.
This article presents two methods for predicting the trajectories of cyclists at an intersection and compares them to a Kalman Filter (KF) approach. The first method uses a physical model of cyclists to predict their future position. The second method is based on a polynomial least-squares approximation in combination with a multilayer perceptron artificial neural network and is able to predict the future position of cyclists independent of their motion type such as “Starting”, “Stopping”, “Waiting” or “Passing”. To evaluate the performance of the methods, 566 tracks (394 for training, 172 for testing) of uninstructed cyclists were recorded at a public intersection using a wide angle stereo camera system and laser scanners. Using the tracks as input data, the future trajectory was predicted for a time horizon of 2.5 s. For starting motions the prediction using the physical model leads to 27% more accurate positions than the KF approach for a forecast horizon of 2.5 s. The neural network shows a 34% more accurate result for starting and stopping motions and a similar result for waiting and passing motions.
In this article, we present an approach for start intention detection of cyclists based on their head trajectories. Therefore, we are using a network architecture based on Long Short-Term Memory (LSTM) cells, which is able to handle input sequences of different lengths. This is important because, for example, due to occlusions, cyclists often only become visible to approaching vehicles shortly before dangerous situations occur. Hence, the dependency of the results on the input sequence length is investigated. We use a dataset with 206 situations where cyclists were transitioning from waiting to moving that was recorded from a moving vehicle in inner-city traffic.With an input sequence length of 1.0 s we achieve an F1-score of 96.2% on average 0.680 s after the first movement of the bicycle. We obtain similar results for sequence lengths down to 0.2 s. For shorter sequences, the results regarding the F1-score and the mean detection time deteriorate considerably.
In this work, we present a new approach for start intention detection of cyclists based on 3D human pose estimation to increase their safety in road traffic. Start intention detection is realized using sequences of frame-wise estimated 3D poses. The poses were obtained by image sequences recorded by a stereo camera mounted behind the windshield of a moving vehicle. For training and evaluation, a dataset with 206 starting cyclists was created in real traffic. We demonstrate the advantages of this approach by comparing it to an existing, solely head trajectory based method. In particular, we investigate the performance for different observed time horizons ranging from 0.12 s up to 1.0 s as inputs for the two methods. This is of special importance for the protection of cyclists in road traffic, as they often only become visible to approaching vehicles shortly before dangerous situations occur. With an input length of 1.0 s the solely head trajectory based approach detects starting motions on average 0.834 s after the first motion of the bicycle with an F1-score of 97.5 %. The pose based approach outperforms these results by achieving the same F1-score 0.135 s earlier. The advantages of the pose based method become even more obvious with shorter input lengths. With an input length of 0.12 s, the head based approach achieves an F1-score of 93.5 % after 2.37 s, while the same score is reached after 0.668 s using poses.
Cooperative Starting Intention Detection of Cyclists based on Smart Devices and Infrastructure
(2017)
In this work, we use Recurrent Neural Networks (RNNs) in form of Gated Recurrent Unit (GRU) networks to forecast trajectories of vulnerable road users (VRUs), such as pedestrians and cyclists, in road traffic utilizing the past trajectory and 3D poses as input. The 3D poses represent the postures and movements of limbs and torso and contain early indicators for the transition between motion types, e.g. wait, start, move, and stop. VRUs often only become visible from the perspective of an approaching vehicle shortly before dangerous situations occur. Therefore, a network architecture is required which is able to forecast trajectories after short time periods and is able to improve the forecasts in case of longer observations. This motivates us to use GRU networks, which are able to use time series of varying duration as inputs, and to investigate the effects of different observation periods on the forecasting results. Our approach is able to make reasonable forecasts even for short observation periods. The use of poses improves the forecasting accuracy, especially for short observation periods compared to a solely head trajectory based approach. Different motion types benefit to different extent from the use of poses and longer observation periods.