Jochen Abhau
Research Subjects
Jochen Abhau has a total of 8 co-authors. Here are the top co-authors by number of shared publications:
- Markus Reischl
Institute for Automation and Applied Informatics (IAI), Karlsruhe Institute of Technology (KIT), Eggenstein-Leopoldshafen, Germany
2 shared publications - Ralf Mikut
Institute for Automation and Applied Informatics (IAI), Karlsruhe Institute of Technology (KIT), Karlsruhe, Germany
2 shared publications - Benjamin Sick
Friedrichshafen AG
1 shared publication - Mark Schutera
Affiliation Institute for Automation and Applied Informatics, Karlsruhe Institute of Technology, Karlsruhe, Baden-Württemberg, Germany
1 shared publication - Mark Schutera
Institute for Automation and Applied Informatics, Karlsruhe Institute of Technology, Eggenstein-Leopoldshafen, Germany
1 shared publication - Michael Walter
Friedrichshafen AG
1 shared publication
Latest Publications
- Open Access
AdaptiveShape: Solving Shape Variability for 3D Object Detection with Geometry Aware Anchor Distributions
3D object detection with point clouds and images plays an important role in perception tasks such as autonomous driving. Current methods show great performance on detection and pose estimation of standard-shaped vehicles but lack behind on more complex shapes as e.g. semi-trailer truck combinations. Determining the shape and motion of those special vehicles accurately is crucial in yard operation and maneuvering and industrial automation applications. This work introduces several new methods to improve and measure the performance for such classes. State-of-the-art methods are based on predefined anchor grids or heatmaps for ground truth targets. However, the underlying representations do not take the shape of different sized objects into account. Our main contribution, AdaptiveShape, uses shape aware anchor distributions and heatmaps to improve the detection capabilities. For large vehicles we achieve +10.9% AP in comparison to current shape agnostic methods. Furthermore we introduce a new fast LiDAR-camera fusion. It is based on 2D bounding box camera detections which are available in many processing pipelines. This fusion method does not rely on perfectly calibrated or temporally synchronized systems and is therefore applicable to a broad range of robotic applications. We extend a standard point pillar network to account for temporal data and improve learning of complex object movements. In addition we extended a ground truth augmentation to use grouped object pairs to further improve truck AP by +2.2% compared to conventional augmentation.
Open AccessAdaptiveShape: Solving Shape Variability for 3D Object Detection with Geometry Aware Anchor Distributions
3D object detection with point clouds and images plays an important role in perception tasks such as autonomous driving. Current methods show great performance on detection and pose estimation of standard-shaped vehicles but lack behind on more complex shapes as e.g. semi-trailer truck combinations. Determining the shape and motion of those special vehicles accurately is crucial in yard operation and maneuvering and industrial automation applications. This work introduces several new methods to improve and measure the performance for such classes. State-of-the-art methods are based on predefined anchor grids or heatmaps for ground truth targets. However, the underlying representations do not take the shape of different sized objects into account. Our main contribution, AdaptiveShape, uses shape aware anchor distributions and heatmaps to improve the detection capabilities. For large vehicles we achieve +10.9% AP in comparison to current shape agnostic methods. Furthermore we introduce a new fast LiDAR-camera fusion. It is based on 2D bounding box camera detections which are available in many processing pipelines. This fusion method does not rely on perfectly calibrated or temporally synchronized systems and is therefore applicable to a broad range of robotic applications. We extend a standard point pillar network to account for temporal data and improve learning of complex object movements. In addition we extended a ground truth augmentation to use grouped object pairs to further improve truck AP by +2.2% compared to conventional augmentation.
- 19 November 2020
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Intelligent Vehicles
- Vol. 6 (3) , 480-489
Night-to-Day: Online Image-to-Image Translation for Object Detection Within Autonomous Driving by Night
Object detectors are central to autonomous driving and widely used in driver assistance systems. Object detectors are trained on a finite amount of data, within a specific domain. This hampers detection performance when applying object detectors to samples from other domains during inference, an effect known as domain gap. Domain gap is a concern for datadriven applications, evoking repetitive retraining of networks when the applications unfold into other domains. With object detectors that have been trained on day images only, domain gap can be clearly observed in object detection by night. Training object detectors on night images is critical due to the enormous effort required to generate an adequate amount of diversely labeled data, and existing data sets often tend to overfit specific domain characteristics. This work proposes, for the first time, adapting domains by online image-to-image translation to expand an object detector's domain of operation. The domain gap is decreased without additional labeling effort, and without having to retrain the object detector, while unfolding into the target domain. The approach follows the concept of domain adaptation, shifting the samples of the target domain into the domain known to the object detector (source domain).
- 19 November 2020
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Intelligent Vehicles
- Vol. 6 (3) , 480-489
Night-to-Day: Online Image-to-Image Translation for Object Detection Within Autonomous Driving by Night
Object detectors are central to autonomous driving and widely used in driver assistance systems. Object detectors are trained on a finite amount of data, within a specific domain. This hampers detection performance when applying object detectors to samples from other domains during inference, an effect known as domain gap. Domain gap is a concern for datadriven applications, evoking repetitive retraining of networks when the applications unfold into other domains. With object detectors that have been trained on day images only, domain gap can be clearly observed in object detection by night. Training object detectors on night images is critical due to the enormous effort required to generate an adequate amount of diversely labeled data, and existing data sets often tend to overfit specific domain characteristics. This work proposes, for the first time, adapting domains by online image-to-image translation to expand an object detector's domain of operation. The domain gap is decreased without additional labeling effort, and without having to retrain the object detector, while unfolding into the target domain. The approach follows the concept of domain adaptation, shifting the samples of the target domain into the domain known to the object detector (source domain).
- 1 July 2019
- Published by Walter de Gruyter GmbH in at - Automatisierungstechnik
- Vol. 67 (7) , 545-556
Strategies for supplementing recurrent neural network training for spatio-temporal prediction
In autonomous driving, prediction tasks address complex spatio-temporal data. This article describes the examination of Recurrent Neural Networks (RNNs) for object trajectory prediction in the image space. The proposed methods enhance the performance and spatio-temporal prediction capabilities of Recurrent Neural Networks. Two different data augmentation strategies and a hyperparameter search are implemented for this purpose. A conventional data augmentation strategy and a Generative Adversarial Network (GAN) based strategy are analyzed with respect to their ability to close the generalization gap of Recurrent Neural Networks. The results are then discussed using single-object tracklets provided by the KITTI Tracking Dataset. This work demonstrates the benefits of augmenting spatio-temporal data with GANs.
- 1 July 2019
- Published by Walter de Gruyter GmbH in at - Automatisierungstechnik
- Vol. 67 (7) , 545-556
Strategies for supplementing recurrent neural network training for spatio-temporal prediction
In autonomous driving, prediction tasks address complex spatio-temporal data. This article describes the examination of Recurrent Neural Networks (RNNs) for object trajectory prediction in the image space. The proposed methods enhance the performance and spatio-temporal prediction capabilities of Recurrent Neural Networks. Two different data augmentation strategies and a hyperparameter search are implemented for this purpose. A conventional data augmentation strategy and a Generative Adversarial Network (GAN) based strategy are analyzed with respect to their ability to close the generalization gap of Recurrent Neural Networks. The results are then discussed using single-object tracklets provided by the KITTI Tracking Dataset. This work demonstrates the benefits of augmenting spatio-temporal data with GANs.