A5000873470- Advanced Image and Video Retrieval Techniques
- Autonomous Vehicle Technology and Safety
- Advanced Chemical Sensor Technologies
- Robotic Path Planning Algorithms
- Multimodal Machine Learning Applications
- Smart Agriculture and AI
- Domain Adaptation and Few-Shot Learning
- Robotics and Sensor-Based Localization
- Advanced Vision and Imaging
- Advanced Manufacturing and Logistics Optimization
- Industrial Vision Systems and Defect Detection
- Target Tracking and Data Fusion in Sensor Networks
- Digital Transformation in Industry
- Anomaly Detection Techniques and Applications
- Advanced Neural Network Applications
- Software Testing and Debugging Techniques
- Gaussian Processes and Bayesian Inference
- Infrared Target Detection Methodologies
- Hand Gesture Recognition Systems
- Advanced Text Analysis Techniques
- Advanced Malware Detection Techniques
- Plant Disease Management Techniques
- Human Pose and Action Recognition
- Fire Detection and Safety Systems
- Reinforcement Learning in Robotics
University of Freiburg
2021-2024
University of Technology Nuremberg
2023
ETH Zurich
2019-2020
Birla Institute of Technology and Science - Hyderabad Campus
2017
Abstract This paper presents the algorithms and system architecture of an autonomous racecar. The introduced vehicle is powered by a software stack designed for robustness, reliability, extensibility. To autonomously race around previously unknown track, proposed solution combines state art techniques from different fields robotics. Specifically, perception, estimation, control are incorporated into one high‐performance complex robotic system, developed AMZ Driverless ETH Zürich, finished...
Bird's-Eye-View (BEV) maps have emerged as one of the most powerful representations for scene understanding due to their ability provide rich spatial context while being easy interpret and process. Such found use in many real-world tasks that extensively rely on accurate segmentation well object instance identification BEV space operation. However, existing algorithms only predict semantics space, which limits applications where notion instances is also critical. In this work, we present...
Bird's-Eye-View (BEV) semantic maps have become an essential component of automated driving pipelines due to the rich representation they provide for decision-making tasks. However, existing approaches generating these still follow a fully supervised training paradigm and hence rely on large amounts annotated BEV data. In this work, we address limitation by proposing first self-supervised approach map using single monocular image from frontal view (FV). During training, overcome need ground...
In autonomous racing, vehicles operate close to the limits of handling and a sensor failure can have critical consequences. To limit impact such failures, this paper presents redundant perception state estimation approaches developed for an race car. Redundancy in is achieved by estimating color position track delimiting objects using two modalities independently. Specifically, learning-based are used generate pose estimates, from LiDAR camera data respectively. The inputs fused particle...
Scene understanding is a pivotal task for autonomous vehicles to safely navigate in the environment. Recent advances deep learning enable accurate semantic reconstruction of surroundings from LiDAR data. However, these models encounter large domain gap while deploying them on equipped with different setups which drastically decreases their performance. Fine-tuning model every new setup infeasible due expensive and cumbersome process recording manually labeling Unsupervised Domain Adaptation...
We present a multi-sensor system for consistent 3D hand pose tracking and modeling that leverages the advantages of both wearable optical sensors. Specifically, we employ stretch-sensing soft glove three IMUs in combination with an RGB-D camera. Different sensor modalities are fused based on availability confidence estimation, enabling seamless challenging environments partial or even complete occlusion. To maximize accuracy while maintaining high ease-of-use, propose automated user...
Perception datasets for agriculture are limited both in quantity and diversity which hinders effective training of supervised learning approaches. Self-supervised techniques alleviate this problem, however, existing methods not optimized dense prediction tasks agricultural domains results degraded performance. In work, we address limitation with our proposed Injected Noise Discriminator (INoD) exploits principles feature replacement dataset discrimination self-supervised representation...
In today's chemical plants, human field operators perform frequent integrity checks to guarantee high safety standards, and thus are possibly the first encounter dangerous operating conditions. To alleviate their task, we present a system consisting of an autonomously navigating robot integrated with various sensors intelligent data processing. It is able detect methane leaks estimate its flow rate, more general gas anomalies, recognize oil films, localize sound sources failure cases, map...
Radar sensors have become an important part of the perception sensor suite due to their long range and ability work in adverse weather conditions. However, several shortcomings such as large amounts noise extreme sparsity point cloud result them not being used full potential. In this paper, we present a novel Recursive Least Squares (RLS) based approach estimate instantaneous velocity dynamic objects real-time that is capable handling input data stream. We also end-to-end pipeline track...
Automated and autonomous industrial inspection is a longstanding research field, driven by the necessity to enhance safety efficiency within settings. In addressing this need, we introduce an autonomously navigating robotic system designed for comprehensive plant inspection. This innovative comprises platform equipped with diverse array of sensors integrated facilitate detection various process infrastructure parameters. These encompass optical (LiDAR, Stereo, UV/IR/RGB cameras), olfactory...
Semantic Bird's Eye View (BEV) maps offer a rich representation with strong occlusion reasoning for various decision making tasks in autonomous driving. However, most BEV mapping approaches employ fully supervised learning paradigm that relies on large amounts of human-annotated ground truth data. In this work, we address limitation by proposing the first unsupervised approach to generate semantic from monocular frontal view (FV) image label-efficient manner. Our pretrains network...
The amalgamation of cloud computing and digital forensics created a new technological domain called forensics. One the technical challenges associated with is Event Reconstruction. reconstruction in significantly differs from traditional environment due to its multi-tenancy large scale events generated per unit time. In this paper, we suggest that effective event possible when number target evidence reduced achieve same by applying aggregation algorithms on service logs. popular log...
Bird's-Eye-View (BEV) semantic maps have become an essential component of automated driving pipelines due to the rich representation they provide for decision-making tasks. However, existing approaches generating these still follow a fully supervised training paradigm and hence rely on large amounts annotated BEV data. In this work, we address limitation by proposing first self-supervised approach map using single monocular image from frontal view (FV). During training, overcome need ground...
Perception datasets for agriculture are limited both in quantity and diversity which hinders effective training of supervised learning approaches. Self-supervised techniques alleviate this problem, however, existing methods not optimized dense prediction tasks domains results degraded performance. In work, we address limitation with our proposed Injected Noise Discriminator (INoD) exploits principles feature replacement dataset discrimination self-supervised representation learning. INoD...
In today's chemical plants, human field operators perform frequent integrity checks to guarantee high safety standards, and thus are possibly the first encounter dangerous operating conditions. To alleviate their task, we present a system consisting of an autonomously navigating robot integrated with various sensors intelligent data processing. It is able detect methane leaks estimate its flow rate, more general gas anomalies, recognize oil films, localize sound sources failure cases, map...
Most automated driving systems comprise a diverse sensor set, including several cameras, Radars, and LiDARs, ensuring complete 360\deg coverage in near far regions. Unlike Radar LiDAR, which measure directly 3D, cameras capture 2D perspective projection with inherent depth ambiguity. However, it is essential to produce perception outputs 3D enable the spatial reasoning of other agents structures for optimal path planning. The space typically simplified BEV by omitting less relevant...
This paper presents the algorithms and system architecture of an autonomous racecar. The introduced vehicle is powered by a software stack designed for robustness, reliability, extensibility. In order to autonomously race around previously unknown track, proposed solution combines state art techniques from different fields robotics. Specifically, perception, estimation, control are incorporated into one high-performance complex robotic system, developed AMZ Driverless ETH Zurich, finished...
Radar sensors have become an important part of the perception sensor suite due to their long range and ability work in adverse weather conditions. However, several shortcomings such as large amounts noise extreme sparsity point cloud result them not being used full potential. In this paper, we present a novel Recursive Least Squares (RLS) based approach estimate instantaneous velocity dynamic objects real-time that is capable handling input data stream. We also end-to-end pipeline track...