UAV Detection with a dataset augmented by domain randomization

Marez, Diego; Borden, Samuel; Nans, Lena

doi:10.1117/12.2558864

Cited by 9 publications

(13 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Researchers can also augment the data set which increases the size of the training data set. Marez et al (2020) augmented their data set using random histogram equalization, horizontal flipping of the image, Gaussian noise, and color jitter. An alternative approach provided by Pedro et al (2020) provides an open repository of 100 videos of drone collisions in different environmental conditions.…”

Section: Safe Transitmentioning

confidence: 99%

Autonomous aerial robotics for package delivery: A technical review

Saunders

Saeedi

2023

Journal of Field Robotics

View full text Add to dashboard Cite

Small unmanned aerial vehicles (UAVs) have gained significant interest in the last decade. More specifically these vehicles have the capacity to impact package delivery logistics in a disruptive way. This paper reviews research problems and state‐of‐the‐art solutions that facilitate package delivery. Different aerial manipulators and grippers are listed along with control techniques to address stability issues. Landing on a platform is next discussed which encompasses static and dynamic platforms. Landing on a dynamic platform presents further challenges. This includes delayed control responses and poor precision of the relative motion between the platform and the aerial vehicle. Subsequently, risks such as weather conditions, state estimation, and collision avoidance to ensure safe transit is considered. Finally, delivery UAV routing is investigated which categorizes the topic into two areas: drone operations and drone–truck collaborative operations. Additionally, we compare the solutions against design, environmental, and legal constraints.

show abstract

Section: Safe Transitmentioning

confidence: 99%

Autonomous aerial robotics for package delivery: A technical review

Saunders

Saeedi

2023

Journal of Field Robotics

View full text Add to dashboard Cite

show abstract

“…Regardless of the application domain, the majority of approaches rely on data synthesis by means of domain randomization (DR) as shown in [13,[20][21][22][23][24][25]27,41]. Based on the idea that non-realistic randomization enhances the learning of essential features, two-or three-dimensional objects of interest (e.g., drones or cars) are typically placed randomly on selected (two-dimensional) backgrounds (e.g., images [23], video sequences [22] or environmental maps [25]).…”

Section: Related Workmentioning

confidence: 99%

“…As data availability remains limited and the cost of acquiring real data remains high, using synthetic data to train DL models represents a popular approach in drone detection [13,[20][21][22][23][24][25] and other application domains (e.g., autonomous driving [26][27][28][29][30]). In addition to their inexpensive generation, utilizing synthetic data offers the potential of overcoming real-world restrictions (e.g., imposed by no-fly zones), enhancing data accessibility, and achieving greater data diversification [31].…”

Section: Introductionmentioning

confidence: 99%

“…However, transferring systems (e.g., object detectors) trained exclusively on simulated data to real-world applications often results in performance degradation caused by the so-called simulation-reality gap (also known as simulation-to-reality gap [36], reality gap [37], domain gap [20], or Sim2Real problem [21,25]). The simulation-reality gap is highly data-dependent and thus directly affected by the quality of the underlying data (both synthetic and real), making its quantification and mitigation a controversial and application-dependent topic of ongoing research [20][21][22][23][26][27][28]30,38].…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Quantifying the Simulation–Reality Gap for Deep Learning-Based Drone Detection

Dieter

Weinmann²,

Jäger

et al. 2023

Electronics

View full text Add to dashboard Cite

The detection of drones or unmanned aerial vehicles is a crucial component in protecting safety-critical infrastructures and maintaining privacy for individuals and organizations. The widespread use of optical sensors for perimeter surveillance has made optical sensors a popular choice for data collection in the context of drone detection. However, efficiently processing the obtained sensor data poses a significant challenge. Even though deep learning-based object detection models have shown promising results, their effectiveness depends on large amounts of annotated training data, which is time consuming and resource intensive to acquire. Therefore, this work investigates the applicability of synthetically generated data obtained through physically realistic simulations based on three-dimensional environments for deep learning-based drone detection. Specifically, we introduce a novel three-dimensional simulation approach built on Unreal Engine and Microsoft AirSim for generating synthetic drone data. Furthermore, we quantify the respective simulation–reality gap and evaluate established techniques for mitigating this gap by systematically exploring different compositions of real and synthetic data. Additionally, we analyze the adaptation of the simulation setup as part of a feedback loop-based training strategy and highlight the benefits of a simulation-based training setup for image-based drone detection, compared to a training strategy relying exclusively on real-world data.

show abstract

“…The use of simulated data along with real data will improve overall detection quality. [23][24][25][26] For cooperative perception between agents most of the datasets rely on the use of gaming engines and simulations. 1,16 Wang et al 1 train V2VNet using simulated LIDAR from Manivasagam et al 19 Following suit with this work, Xu et al 27 have generated a publicly available platform where multiple cars can navigate one scene and generate data.…”

Section: Datasetsmentioning

confidence: 99%

Bandwidth constrained cooperative object detection in images

Marez

Nans²,

Borden³

2022

Artificial Intelligence and Machine Learning in Defense Applications IV

View full text Add to dashboard Cite

Cooperative autonomous systems, such as swarms, multi-camera systems, or fleets of self-driving vehicles, can better understand a given scene by sharing multiple modalities and varying viewpoints. This superposition of data adds robustness and redundancy to a system typically burdened with obstructions and unrecognizable, distant objects. Collaborative perception is a key component of cooperative autonomous systems where modalities can include camera sensors, LiDAR, RADAR, and depth images. Meanwhile, the amount of useful information that can be shared between agents in a cooperative system is constrained by current communication technologies (e.g. bandwidth limitations). Recent developments in learned compression can enable the training of end-toend cooperative systems using deep learning with compressed communication in the pipeline. We explore the use of a deep learning object detector in a cooperative setting with a learned compression model facilitating communication between agents. To test our algorithm, this research will focus on object detection in the image domain as a proxy for one of the modalities used by collaborative systems.

show abstract

UAV Detection with a dataset augmented by domain randomization

Cited by 9 publications

References 0 publications

Autonomous aerial robotics for package delivery: A technical review

Autonomous aerial robotics for package delivery: A technical review

Quantifying the Simulation–Reality Gap for Deep Learning-Based Drone Detection

Bandwidth constrained cooperative object detection in images

Contact Info

Product

Resources

About