Search

The best performing learning algorithms devised for event cameras work by first converting events into dense representations that are then processed using standard CNNs. However, these steps discard both the sparsity and high temporal resolution of events, leading to high computational burden and latency. For this reason, recent works have adopted Graph Neural Networks (GNNs), which process events as ``static" spatio-temporal graphs, which are inherently "sparse". We take this trend one step further by introducing Asynchronous, Event-based Graph Neural Networks (AEGNNs), a novel event-processing paradigm that generalizes standard GNNs to process events as ``evolving" spatio-temporal graphs. AEGNNs follow efficient update rules that restrict recomputation of network activations only to the nodes affected by each new event, thereby significantly reducing both computation and latency for event-by-event processing. AEGNNs are easily trained on synchronous inputs and can be converted to efficient, "asynchronous" networks at test time. We thoroughly validate our method on object classification and detection tasks, where we show an up to a 11-fold reduction in computational complexity (FLOPs), with similar or even better performance than state-of-the-art asynchronous methods. This reduction in computation directly translates to an 8-fold reduction in computational latency when compared to standard GNNs, which opens the door to low-latency event-based processing., IEEE Conference on Computer Vision and Pattern Recognition (CVPR), New Orleans, 2022

Event cameras are novel sensors that output brightness changes in the form of a stream of asynchronous "events" instead of intensity frames. They offer significant advantages with respect to conventional cameras: high dynamic range (HDR), high temporal resolution, and no motion blur. Recently, novel learning approaches operating on event data have achieved impressive results. Yet, these methods require a large amount of event data for training, which is hardly available due the novelty of event sensors in computer vision research. In this paper, we present a method that addresses these needs by converting any existing video dataset recorded with conventional cameras to synthetic event data. This unlocks the use of a virtually unlimited number of existing video datasets for training networks designed for real event data. We evaluate our method on two relevant vision tasks, i.e., object recognition and semantic segmentation, and show that models trained on synthetic events have several benefits: (i) they generalize well to real event data, even in scenarios where standard-camera images are blurry or overexposed, by inheriting the outstanding properties of event cameras; (ii) they can be used for fine-tuning on real data to improve over state-of-the-art for both classification and semantic segmentation.

State-of-the-art frame interpolation methods generate intermediate frames by inferring object motions in the image from consecutive key-frames. In the absence of additional information, first-order approximations, i.e. optical flow, must be used, but this choice restricts the types of motions that can be modeled, leading to errors in highly dynamic scenarios. Event cameras are novel sensors that address this limitation by providing auxiliary visual information in the blind-time between frames. They asynchronously measure per-pixel brightness changes and do this with high temporal resolution and low latency. Event-based frame interpolation methods typically adopt a synthesis-based approach, where predicted frame residuals are directly applied to the key-frames. However, while these approaches can capture non-linear motions they suffer from ghosting and perform poorly in low-texture regions with few events. Thus, synthesis-based and flow-based approaches are complementary. In this work, we introduce Time Lens, a novel method that leverages the advantages of both. We extensively evaluate our method on three synthetic and two real benchmarks where we show an up to 5.21 dB improvement in terms of PSNR over state-of-the-art frame-based and event-based methods. Finally, we release a new large-scale dataset in highly dynamic scenarios, aimed at pushing the limits of existing methods.

Event cameras are novel sensors that output brightness changes in the form of a stream of asynchronous events instead of intensity frames. Compared to conventional image sensors, they offer significant advantages: high temporal resolution, high dynamic range, no motion blur, and much lower bandwidth. Recently, learning-based approaches have been applied to event-based data, thus unlocking their potential and making significant progress in a variety of tasks, such as monocular depth prediction. Most existing approaches use standard feed-forward architectures to generate network predictions, which do not leverage the temporal consistency presents in the event stream. We propose a recurrent architecture to solve this task and show significant improvement over standard feed-forward methods. In particular, our method generates dense depth predictions using a monocular setup, which has not been shown previously. We pretrain our model using a new dataset containing events and depth maps recorded in the CARLA simulator. We test our method on the Multi Vehicle Stereo Event Camera Dataset (MVSEC). Quantitative experiments show up to 50% improvement in average depth error with respect to previous event-based methods., IEEE International Conference on 3D Vision (3DV), 2020

Event cameras are vision sensors that record asynchronous streams of per-pixel brightness changes, referred to as "events". They have appealing advantages over frame-based cameras for computer vision, including high temporal resolution, high dynamic range, and no motion blur. Due to the sparse, non-uniform spatiotemporal layout of the event signal, pattern recognition algorithms typically aggregate events into a grid-based representation and subsequently process it by a standard vision pipeline, e.g., Convolutional Neural Network (CNN). In this work, we introduce a general framework to convert event streams into grid-based representations through a sequence of differentiable operations. Our framework comes with two main advantages: (i) allows learning the input event representation together with the task dedicated network in an end to end manner, and (ii) lays out a taxonomy that unifies the majority of extant event representations in the literature and identifies novel ones. Empirically, we show that our approach to learning the event representation end-to-end yields an improvement of approximately 12% on optical flow estimation and object recognition over state-of-the-art methods., To appear at ICCV 2019

We propose a novel scoring concept for visual place recognition based on nearest neighbor descriptor voting and demonstrate how the algorithm naturally emerges from the problem formulation. Based on the observation that the number of votes for matching places can be evaluated using a binomial distribution model, loop closures can be detected with high precision. By casting the problem into a probabilistic framework, we not only remove the need for commonly employed heuristic parameters but also provide a powerful score to classify matching and non-matching places. We present methods for both a 2D-2D image matching and a 2D-3D landmark matching based on the above scoring. The approach maintains accuracy while being efficient enough for online application through the use of compact (low-dimensional) descriptors and fast nearest neighbor retrieval techniques. The proposed methods are evaluated on several challenging datasets in varied environments, showing state-of-the-art results with high precision and high recall.

Advanced Driver Assistance Systems benefit from a full 3D reconstruction of the environment in real-time, often obtained via stereo vision. Semi-Global Matching (SGM) is a popular stereo algorithm for solving this task which is already in use for production vehicles. Despite this progess, one key challenge remains: stereo vision during adverse weather conditions such as rain, snow and low-lighting. Current methods generate many disparity outliers and false positives on a segmentation level under such conditions. These shortcomings are alleviated by integrating prior scene knowledge. We formulate a scene prior that exploits knowledge of a representative traffic scene, which we apply to SGM and Graph Cut based disparity estimation. The prior is learned from traffic scene statistics extracted during good weather. Using this prior, the object detection rate is maintained on a driver assistance database of 3000 frames including bad weather while reducing the false positive rate significantly. Similar results are obtained for the KITTI dataset, maintaining excellent performance in good weather conditions. We also show that this scene prior is easy and efficient to implement both on CPU platforms and on reconfigurable hardware platforms. The concept can be extended to other application areas such as indoor robotics, when prior information of the disparity distribution is gathered.

Autonomous Driving benefits strongly from a 3D reconstruction of the environment in real-time, often obtained via stereo vision. Semi-Global Matching (SGM) is a popular method of choice for solving this task which is already in use for production vehicles. Despite the enormous progress in the field and the high level of performance of modern methods, one key challenge remains: stereo vision in automotive scenarios during weather conditions such as rain, snow and night. Current methods generate strong temporal noise, many disparity outliers and false positives on a segmentation level. They are addressed in this work. We formulate a temporal prior and a scene prior, which we apply to SGM and Graph Cut. Using these priors, the object detection rate improves significantly on a driver assistance database of 3000 frames including bad weather while reducing the false positive rate. We also outperform the ECCV Robust Vision Challenge winner, iSGM, on this database.

Applications based on stereo vision are becoming increasingly common, ranging from gaming over robotics to driver assistance. While stereo algorithms have been investigated heavily both on the pixel and the application level, far less attention has been dedicated to the use of stereo confidence cues. Mostly, a threshold is applied to the confidence values for further processing, which is essentially a sparsified disparity map. This is straightforward but it does not take full advantage of the available information. In this paper, we make full use of the stereo confidence cues by propagating all confidence values along with the measured disparities in a Bayesian manner. Before using this information, a mapping from confidence values to disparity outlier probability rate is performed based on gathered disparity statistics from labeled video data. We present an extension of the so called Stixel World, a generic 3D intermediate representation that can serve as input for many of the applications mentioned above. This scheme is modified to directly exploit stereo confidence cues in the underlying sensor model during a maximum a posteriori estimation process. The effectiveness of this step is verified in an in-depth evaluation on a large real-world traffic data base of which parts are made publicly available. We show that using stereo confidence cues allows both reducing the number of false object detections by a factor of six while keeping the detection rate at a near constant level.

In this contribution we present an approach to compute object confidences for stereo-vision-based object tracking schemes. Meaningful object confidences help to reduce false alarm rates of safety systems and improve the downstream system performance for modules such as sensor fusion and situation analysis. Several cues from stereo vision and from the tracking process are fused in a Bayesian manner. An evaluation on a 38,000 frames urban drive shows the effectiveness of the approach compared to the same object tracking scheme with simple heuristics for the object confidence. Within the evaluation, also the relevance of occurring phantoms is considered by computing the collision risk. The proposed confidence measures reduce the number of predicted imminent collisions from 86 to 0 maintaining almost the same system availability.

We examine high accuracy stereo estimation for binocular sequences that where obtained from a mobile platform. The ultimate goal is to improve the range of stereo systems without altering the setup. Based on a well-known variational optical flow model, we introduce a novel stereo model that features a second-order regularization, which both allows sub-pixel accurate solutions and piecewise planar disparity maps. The model incorporates a robust fidelity term to account for adverse illumination conditions that frequently arise in real-world scenes. Using several sequences that were taken from a mobile platform we show the robustness and accuracy of the proposed model.

In recent years, many dense optical flow algorithms have been presented. Besides the actual result, the confidence of the result is of uttermost importance for safety-critical applications using flow. We focus on the determination of the confidence of optical flow.

In this work, we propose a framework for simultaneously detecting the presence of multiple facial action units using kernel partial least square regression (KPLS). This method has the advantage of being easily extensible to learn more face related labels, while at the same time being computationally efficient. We compare the approach to linear and non-linear support vector machines (SVM) and evaluate its performance on the extended Cohn-Kanade (CK+) dataset and the GEneva Multimodal Emotion Portrayals (GEMEP-FERA) dataset, as well as across databases. It is shown that KPLS achieves around 2% absolute improvement over the SVM-based approach in terms of the two alternative forced choice (2AFC) score when trained on CK+ and tested on CK+ and GEMEP-FERA. It achieves around 6% absolute improvement over the SVM-based approach when trained on GEMEP-FERA and tested on CK+. We also show that KPLS is handling non-additive AU combinations better than SVM-based approaches trained to detect single AUs only.

In this paper, a multi-level approach to intention, activity, and motion recognition for a humanoid robot is proposed. Our system processes images from a monocular camera and combines this information with domain knowledge. The recognition works on-line and in real-time, it is independent of the test person, but limited to predefined view-points. Main contributions of this paper are the extensible, multi-level modeling of the robot's vision system, the efficient activity and motion recognition, and the asynchronous information fusion based on generic processing of mid-level recognition results. The complementarity of the activity and motion recognition renders the approach robust against misclassifications. Experimental results on a real-world data set of complex kitchen tasks, e.g., Prepare Cereals or Lay Table, prove the performance and robustness of the multi-level recognition approach.