Given a ground-level query image and a geo-referenced aerial image that covers the query’s local surroundings, fine-grained cross-view localization aims to estimate the location of the ground camera inside the aerial image. Recent works have focused on developing advanced networks trained with accurate ground truth (GT) locations of ground images. However, the trained models always suffer a performance drop when applied to images in a new target area that differs from training. In most deployment scenarios, acquiring fine GT, i.e. accurate GT locations, for target-area images to re-train the network can be expensive and sometimes infeasible. In contrast, collecting images with noisy GT with errors of tens of meters is often easy. Motivated by this, our paper focuses on improving the performance of a trained model in a new target area by leveraging only the target-area images without fine GT. We propose a weakly supervised learning approach based on knowledge self-distillation. This approach uses predictions from a pre-trained model as pseudo GT to supervise a copy of itself. Our approach includes a mode-based pseudo GT generation for reducing uncertainty in pseudo GT and an outlier filtering method to remove unreliable pseudo GT. Our approach is validated using two recent state-of-the-art models on two benchmarks. The results demonstrate that it consistently and considerably boosts the localization accuracy in the target area.@en

We introduce MuVieCAST, a modular multi-view consistent style transfer network architecture that enables consistent style transfer between multiple viewpoints of the same scene. This network architecture supports both sparse and dense views, making it versatile enough to handle a wide range of multi-view image datasets. The approach consists of three modules that perform specific tasks related to style transfer, namely content preservation, image transformation, and multi-view consistency enforcement. We extensively evaluate our approach across multiple application domains including depth-map-based point cloud fusion, mesh reconstruction, and novel-view synthesis. Our experiments reveal that the proposed framework achieves an exceptional generation of stylized images, exhibiting consistent outcomes across perspectives. A user study focusing on novel-view synthesis further confirms these results, with approximately 68% of cases participants expressing a preference for our generated outputs compared to the recent state-of-the-art method. Our modular framework is extensible and can easily be integrated with various backbone architectures, making it a flexible solution for multi-view style transfer.@en

A substantial number of vehicles nowadays are equipped with adaptive cruise control (ACC), which adjusts the vehicle speed automatically. However, experiments have found that commercial ACC systems which only detect the direct leader amplify the propagating disturbances in the platoon. This can cause severe traffic congestion when the number of ACC-equipped vehicles increases. Therefore, an ACC system which also considers the second leader further downstream is required. Such a system enables the vehicle to achieve multi-anticipation and hence ensure better platoon stability. Nevertheless, measurements collected from the second leader may be comparatively inaccurate given the limitations of current state-of-the-art sensor technology. This study adopts deep reinforcement learning to develop ACC controllers that besides the input from the first leader exploits the additional information obtained from the second leader, albeit noisy. The simulation experiment demonstrates that even under the influence of noisy measurements, the multi-leader ACC platoon shows smaller disturbance and jerk amplitudes than the one-leader ACC platoon, indicating improved string stability and ride comfort. Practical takeaways are twofold: first, the proposed method can be used to further develop multi-leader ACC systems. Second, even noisy data from the second leader can help stabilize traffic, which makes such systems viable in practice.

@en

AI in automotive is sterk in opkomst. Nieuwe voertuigen zijn always connected en worden zelflerende machines. Zo pakt de techniek een steeds grotere rol in de bediening en besturing van onze voertuigen. Wat betekent dat voor de veiligheid? En verandert dat ook de rol van de RDW?@en

In this paper, we address the limitations of traditional constant false alarm rate (CFAR) target detectors in automotive radars, particularly in complex urban environments with multiple objects that appear as extended targets. We propose a data-driven radar target detector exploiting a highly efficient 2D CNN backbone inspired by the computer vision domain. Our approach is distinguished by a unique cross-sensor supervision pipeline, enabling it to learn exclusively from unlabeled synchronized radar and lidar data, thus^eliminating the need for costly manual object annotations. Using a novel large-scale, real-life multi-sensor dataset recorded in various driving scenarios, we demonstrate that the proposed detector generates dense, lidar-like point clouds, achieving a lower Chamfer distance to the reference lidar point clouds than CFAR detectors. Overall, it significantly outperforms CFAR baselines detection accuracy.

@en

We propose a novel end-to-end method for cross-view pose estimation. Given a ground-level query image and an aerial image that covers the query's local neighborhood, the 3 Degrees-of-Freedom camera pose of the query is estimated by matching its image descriptor to descriptors of local regions within the aerial image. The orientation-aware descriptors are obtained by using a translationally equivariant convolutional ground image encoder and contrastive learning. The Localization Decoder produces a dense probability distribution in a coarse-to-fine manner with a novel Localization Matching Upsampling module. A smaller Orientation Decoder produces a vector field to condition the orientation estimate on the localization. Our method is validated on the VIGOR and KITTI datasets, where it surpasses the state-of-the-art baseline by 72% and 36% in median localization error for comparable orientation estimation accuracy. The predicted probability distribution can represent localization ambiguity, and enables rejecting possible erroneous predictions. Without re-training, the model can infer on ground images with different field of views and utilize orientation priors if available. On the Oxford RobotCar dataset, our method can reliably estimate the ego-vehicle's pose over time, achieving a median localization error under 1 meter and a median orientation error of around 1 degree at 14 FPS.

@en

This letter presents View-of-Delft Prediction, a new dataset for trajectory prediction, to address the lack of on-board trajectory datasets in urban mixed-traffic environments. View-of-Delft Prediction builds on the recently released urban View-of-Delft (VoD) dataset to make it suitable for trajectory prediction. Unique features of this dataset are the challenging road layouts of Delft, with many narrow roads and bridges, and the close proximity between vehicles and Vulnerable Road Users (VRUs). It contains a large proportion of VRUs, with 569 prediction instances for vehicles, 347 for cyclists, and 934 for pedestrians. We additionally provide high-definition map annotations for the VoD dataset to enable state-of-the-art prediction models to be used. We analyse two state-of-the-art trajectory prediction models, PGP and P2T, which originally were developed for vehicle-dominated traffic scenarios, to assess the strengths and weaknesses of current modelling approaches in mixed traffic settings with large numbers of VRUs. Our analysis shows that there is a significant domain gap between the vehicle-dominated nuScenes and VRU-dominated VoD Prediction datasets. The dataset is publicly released for non-commercial research purposes.

@en

Multi-sensor object detection is an active research topic in automated driving, but the robustness of such detection models against missing sensor input (modality missing), e.g., due to a sudden sensor failure, is a critical problem which remains under-studied. In this work, we propose UniBEV, an end-to-end multi-modal 3D object detection framework designed for robustness against missing modalities: UniBEV can operate on LiDAR plus camera input, but also on LiDAR-only or camera-only input without retraining. To facilitate its detector head to handle different input combinations, UniBEV aims to create well-aligned Bird’s Eye View (BEV) feature maps from each available modality. Unlike prior BEV-based multi-modal detection methods, all sensor modalities follow a uniform approach to resample features from the original sensor coordinate systems to the BEV features. We furthermore investigate the robustness of various fusion strategies w.r.t. missing modalities: the commonly used feature concatenation, but also channel-wise averaging, and a generalization to weighted averaging termed Channel Normalized Weights. To validate its effectiveness, we compare UniBEV to state-of-the-art BEVFusion and MetaBEV on nuScenes over all sensor input combinations. In this setting, UniBEV achieves better performance than these baselines for all input combinations. An ablation study shows the robustness benefits of fusing by weighted averaging over regular concatenation, and of sharing queries between the BEV encoders of each modality. Our code is available at https://github.com/tudelft-iv/UniBEV.@en

This work addresses cross-view camera pose estimation, i.e., determining the 3-Degrees-of-Freedom camera pose of a given ground-level image w.r.t. an aerial image of the local area. We propose SliceMatch, which consists of ground and aerial feature extractors, feature aggregators, and a pose predictor. The feature extractors extract dense features from the ground and aerial images. Given a set of candidate camera poses, the feature aggregators construct a single ground descriptor and a set of pose-dependent aerial descriptors. Notably, our novel aerial feature aggregator has a cross-view attention module for ground-view guided aerial feature selection and utilizes the geometric projection of the ground camera's viewing frustum on the aerial image to pool features. The efficient construction of aerial descriptors is achieved using precomputed masks. SliceMatch is trained using contrastive learning and pose estimation is formulated as a similarity comparison between the ground descriptor and the aerial descriptors. Compared to the state-of-the-art, SliceMatch achieves a 19% lower median localization error on the VIGOR benchmark using the same VGG16 backbone at 150 frames per second, and a 50% lower error when using a ResNet50 backbone.@en

We propose an enhancement module called depth discontinuity learning (DDL) for learning-based multi-view stereo (MVS) methods. Traditional methods are known for their accuracy but struggle with completeness. While recent learning-based methods have improved completeness at the cost of accuracy, our DDL approach aims to improve accuracy while retaining completeness in the reconstruction process. To achieve this, we introduce the joint estimation of depth and boundary maps, where the boundary maps are explicitly utilized for further refinement of the depth maps. We validate our idea by integrating it into an existing learning-based MVS pipeline where the reconstruction depends on high-quality depth map estimation. Extensive experiments on various datasets, namely DTU, ETH3D, “Tanks and Temples”, and BlendedMVS, show that our method improves reconstruction quality compared to our baseline, Patchmatchnet. Our ablation study demonstrates that incorporating the proposed DDL significantly reduces the depth map error, for instance, by more than 30% on the DTU dataset, and leads to improved depth map quality in both smooth and boundary regions. Additionally, our qualitative analysis has shown that the reconstructed point cloud exhibits enhanced quality without any significant compromise on completeness. Finally, the experiments reveal that our proposed model and strategies exhibit strong generalization capabilities across the various datasets.@en

Visual place recognition (VPR) is an image-based localization method that estimates the camera location of a query image by retrieving the most similar reference image from a map of geo-tagged reference images. In this work, we look into two fundamental bottlenecks for its localization accuracy: 1) reference map sparseness and 2) viewpoint invariance. First, the reference images for VPR are only available at sparse poses in a map, which enforces an upper bound on the maximum achievable localization accuracy through VPR. We, therefore, propose Continuous Place-descriptor Regression (CoPR) to densify the map and improve localization accuracy. We study various interpolation and extrapolation models to regress additional VPR feature descriptors from only the existing references. Second, we compare different feature encoders and show that CoPR presents value for all of them. We evaluate our models on three existing public datasets and report on average around 30% improvement in VPR-based localization accuracy using CoPR, on top of the 15% increase by using a viewpoint-variant loss for the feature encoder. The complementary relation between CoPR and relative pose estimation is also discussed.@en

Early and accurate detection of crossing pedestrians is crucial in automated driving in order to perform timely emergency manoeuvres. However, this is a difficult task in urban scenarios where pedestrians are often occluded (not visible) behind objects, e.g., other parked vehicles. We propose an occlusion aware fusion of stereo camera and radar sensors to address scenarios with crossing pedestrians behind such parked vehicles. Our proposed method adapts both the expected rate and properties of detections in different areas according to the visibility of the sensors. In our experiments on a real-world dataset, we show that the proposed occlusion aware fusion of radar and stereo camera detects the crossing pedestrians on average 0.26 seconds earlier than using the camera alone, and 0.15 seconds earlier than fusing the sensors without occlusion information. Our dataset containing 501 relevant recordings of pedestrians behind vehicles will be publicly available on our website for non-commercial, scientific use.

@en

This work addresses visual cross-view metric localization for outdoor robotics. Given a ground-level color image and a satellite patch that contains the local surroundings, the task is to identify the location of the ground camera within the satellite patch. Related work addressed this task for range-sensors (LiDAR, Radar), but for vision, only as a secondary regression step after an initial cross-view image retrieval step. Since the local satellite patch could also be retrieved through any rough localization prior (e.g. from GPS/GNSS, temporal filtering), we drop the image retrieval objective and focus on the metric localization only. We devise a novel network architecture with denser satellite descriptors, similarity matching at the bottleneck (rather than at the output as in image retrieval), and a dense spatial distribution as output to capture multi-modal localization ambiguities. We compare against a state-of-the-art regression baseline that uses global image descriptors. Quantitative and qualitative experimental results on the recently proposed VIGOR and the Oxford RobotCar datasets validate our design. The produced probabilities are correlated with localization accuracy, and can even be used to roughly estimate the ground camera’s heading when its orientation is unknown. Overall, our method reduces the median metric localization error by 51%, 37%, and 28% compared to the state-of-the-art when generalizing respectively in the same area, across areas, and across time.

@en

Customization of a convolutional neural network (CNN) to a specific compute platform involves finding an optimal pareto state between computational complexity of the CNN and resulting throughput in operations per second on the compute platform. However, existing inference performance benchmarks compare complete backbones that entail many differences between their CNN configurations, which do not provide insights in how fine-grade layer design choices affect this balance.We present BackboneAnalysis, a methodology for extracting structured insights into the trade-off for a chosen target compute platform. Within a one-factor-at-a-time analysis setup, CNN architectures are systematically varied and evaluated based on throughput and latency measurements irrespective of model accuracy. Thereby, we investigate the configuration factors input shape, batch size, kernel size and convolutional layer type.In our experiments, we deploy BackboneAnalysis on a Xavier iGPU and a Coral Edge TPU accelerator. The analysis reveals that the general assumption from optimal Roofline performance that higher operation density in CNNs leads to higher throughput does not always hold. These results highlight the importance for a neural network architect to be aware of platform-specific latency and throughput behavior in order to derive sensible configuration decisions for a custom CNN.@en

Next-generation automotive radars provide elevation data in addition to range-, azimuth- and Doppler velocity. In this experimental study, we apply a state-of-the-art object detector (PointPillars), previously used for LiDAR 3D data, to such 3+1D radar data (where 1D refers to Doppler). In ablation studies, we first explore the benefits of the additional elevation information, together with that of Doppler, radar cross section and temporal accumulation, in the context of multi-class road user detection. We subsequently compare object detection performance on the radar and LiDAR point clouds, object class-wise and as a function of distance. To facilitate our experimental study, we present the novel View-of-Delft (VoD) automotive dataset. It contains 8693 frames of synchronized and calibrated 64-layer LiDAR-, (stereo) camera-, and 3+1D radar-data acquired in complex, urban traffic. It consists of 123106 3D bounding box annotations of both moving and static objects, including 26587 pedestrian, 10800 cyclist and 26949 car labels. Our results show that object detection on 64-layer LiDAR data still outperforms that on 3+1D radar data, but the addition of elevation information and integration of successive radar scans helps close the gap. The VoD dataset is made freely available for scientific benchmarking.

@en

Feedforward fully convolutional neural networks currently dominate in semantic segmentation of 3D point clouds. Despite their great success, they suffer from the loss of local information at low-level layers, posing significant challenges to accurate scene segmentation and precise object boundary delineation. Prior works either address this issue by post-processing or jointly learn object boundaries to implicitly improve feature encoding of the networks. These approaches often require additional modules which are difficult to integrate into the original architecture. To improve the segmentation near object boundaries, we propose a boundary-aware feature propagation mechanism. This mechanism is achieved by exploiting a multitask learning framework that aims to explicitly guide the boundaries to their original locations. With one shared encoder, our network outputs (i) boundary localization, (ii) prediction of directions pointing to the object's interior, and (iii) semantic segmentation, in three parallel streams. The predicted boundaries and directions are fused to propagate the learned features to refine the segmentation. We conduct extensive experiments on the S3DIS and SensatUrban datasets against various baseline methods, demonstrating that our proposed approach yields consistent improvements by reducing boundary errors. Our code is available at https://github.com/shenglandu/PushBoundary.

@en

Person re-identification is a key challenge for surveillance across multiple sensors. Prompted by the advent of powerful deep learning models for visual recognition, and inexpensive RGB-D cameras and sensor-rich mobile robotic platforms, e.g. self-driving vehicles, we investigate the relatively unexplored problem of cross-modal re-identification of persons between RGB (color) and depth images. The considerable divergence in data distributions across different sensor modalities introduces additional challenges to the typical difficulties like distinct viewpoints, occlusions, and pose and illumination variation. While some work has investigated re-identification across RGB and infrared, we take inspiration from successes in transfer learning from RGB to depth in object detection tasks. Our main contribution is a novel method for cross-modal distillation for robust person re-identification, which learns a shared feature representation space of person's appearance in both RGB and depth images. In addition, we propose a cross-modal attention mechanism where the gating signal from one modality can dynamically activate the most discriminant CNN filters of the other modality. The proposed distillation method is compared to conventional and deep learning approaches proposed for other cross-domain re-identification tasks. Results obtained on the public BIWI and RobotPKU datasets indicate that the proposed method can significantly outperform the state-of-the-art approaches by up to 16.1% in mean Average Precision (mAP), demonstrating the benefit of the distillation paradigm. The experimental results also indicate that using cross-modal attention allows to improve recognition accuracy considerably with respect to the proposed distillation method and relevant state-of-the-art approaches.¹

@en

State-of-the-art stixel methods fuse dense stereo disparity and semantic class information, e.g. from a Convolutional Neural Network (CNN), into a compact representation of driveable space, obstacles and background. However, they do not explicitly differentiate instances within the same semantic class. We investigate several ways to augment single-frame stixels with instance information, which can be extracted by a CNN from the RGB image input. As a result, our novel Instance Stixels method efficiently computes stixels that account for boundaries of individual objects, and represents instances as grouped stixels that express connectivity. Experiments on the Cityscapes dataset demonstrate that including instance information into the stixel computation itself, rather than as a post-processing step, increases the segmentation performance (i.e. Intersection over Union and Average Precision). This holds especially for overlapping objects of the same class. Furthermore, we show the superiority of our approach in terms of segmentation performance and computational efficiency compared to combining the separate outputs of Semantic Stixels and a state-of-the-art pixel-level CNN. We achieve processing throughput of 28 frames per second on average for 8 pixel wide stixels on images from the Cityscapes dataset at 1792x784 pixels. Our Instance Stixels software is made freely available for non-commercial research purposes.

@en

Cross-view matching aims to learn a shared image representation between ground-level images and satellite or aerial images at the same locations. In robotic vehicles, matching a camera image to a database of geo-referenced aerial imagery can serve as a method for self-localization. However, existing work on cross-view matching only aims at global localization, and overlooks the easily accessible rough location estimates from GNSS or temporal filtering. We argue that the availability of coarse location estimates at test time should already be considered during training. We adopt a simple but effective adaptation to the common triplet loss, resulting in an image representation that is more discriminative within the geographically local neighborhood, without any modifications to a baseline deep neural network. Experiments on the CVACT dataset confirm that the improvements generalize across spatial regions. On a new benchmark constructed from the Oxford RobotCar dataset, we also show generalization across recording days within the same region. Finally, we validate that improvements on these image-retrieval benchmarks also translate to a real-world localization task. Using a particle filter to fuse the cross-view matching scores of a vehicle's camera stream with real GPS measurements, our learned geographically local representation reduces the mean localization error by 17\% compared to the standard global representation learned by the current state-of-the-art.

@en

We address joint extrinsic calibration of lidar, camera and radar sensors. To simplify calibration, we propose a single calibration target design for all three modalities, and implement our approach in an open-source tool with bindings to Robot Operating System (ROS). Our tool features three optimization configurations, namely using error terms for a minimal number of sensor pairs, or using terms for all sensor pairs in combination with loop closure constraints, or by adding terms for structure estimation in a probabilistic model. Apart from relative calibration where relative transformations between sensors are computed, our work also addresses absolute calibration that includes calibration with respect to the mobile robot's body. Two methods are compared to estimate the body reference frame using an external laser scanner, one based on markers and the other based on manual annotation of the laser scan. In the experiments, we evaluate the three configurations for relative calibration. Our results show that using terms for all sensor pairs is most robust, especially for lidar to radar, when minimum five board locations are used. For absolute calibration the median rotation error around the vertical axis reduces from 1. before calibration, to 0.33. using the markers and 0.02. with manual annotations.

@en