Super accurate low latency object detection on a surveillance UAV

Vandersteegen, Maarten; Beeck, Kristof Van; Goedemé, Toon

doi:10.23919/mva.2019.8758060

Cited by 15 publications

(8 citation statements)

References 22 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Sommer et al [25] improved the detection performance of small objects in aerial images by modifying the network structure and deploying data augmentation. Vandersteegen et al [45] optimized anchor size and tested different pretraining models to improve aerial images' detection accuracy.…”

Section: E Quantitative Resultsmentioning

confidence: 99%

Toward Efficient Object Detection in Aerial Images Using Extreme Scale Metric Learning

Ren

et al. 2021

IEEE Access

View full text Add to dashboard Cite

In aerial image object detection, how to efficiently detect different size objects in input images of different scales and obtain a unified multi-scale representation of the object is an important issue. Existing methods rarely consider the connection between multi-scale training and multi-scale inference, and do not well optimize the constraint of input object samples in the multi-scale training process, which limits the performance of multi-scale representation. In this study, an efficient object detection algorithm for aerial images is proposed to alleviate this problem. Firstly, we propose to use metric learning to obtain the scale representation boundary of each object class, reduce the support of indistinguishable objects at extreme scales in the training process, and enhance the effect of multi-scale representation. Secondly, indistinguishable small objects are merged into small object regions, and these regions are trained to recommend the detector to detect small objects on the following high-resolution scale. Thus, a reasonable association between multi-scale training and inference is established, and the efficiency of multiscale inference is considerably improved. The proposed algorithm has been tested on three popular aerial image datasets, including VisDrone, DOTA and UAVDT. Experimental results show that it can improve the detection accuracy and reduce the number of processing pixels.INDEX TERMS aerial images, extreme scale, metric learning, object detection.

show abstract

Section: E Quantitative Resultsmentioning

confidence: 99%

Toward Efficient Object Detection in Aerial Images Using Extreme Scale Metric Learning

Ren

et al. 2021

IEEE Access

View full text Add to dashboard Cite

show abstract

“…This arrangement enables answering the question of whether the specified region is suitable for attachment using teeth; if the answer is 'no' then the vacuum suction mode is deployed, which is associated with a probability value of less than 56% in our design. Note that, the main reason for choosing the YOLOv3 algorithm was the ability to run this type of algorithm on NVIDIA ® Jetson Xavier ™ NX with low latency (virtually real-time) at a high precision [25]. In order to design and implement the algorithm a dataset of 500 images from environments that are suitable for attachment via spiky teeth is created via photography, then all images within the dataset are labeled.…”

Section: Visual Examination Of Surface Propertiesmentioning

confidence: 99%

Autonomous Decision Making in a Bioinspired Adaptive Robotic Anchoring Module

Sadeghian

Sareh

Shahin

et al. 2021

2021 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)

View full text Add to dashboard Cite

This paper proposes a bioinspired adaptive anchoring module that can be integrated into robots to enhance their mobility and manipulation abilities. The design of the module is inspired by the structure of the mouth in Chilean lamprey (Mordacia lapicida) where a combination of suction and several arrays of teeth with different sizes around the mouth opening is used for catching preys and anchoring onto them. The module can deploy a suitable mode of attachment, via teeth or vacuum suction, to different contact surfaces in response to the textural properties of those surfaces. In order to make a decision on the suitable mode of attachment, an original dataset of 500 images of outdoor and indoor surfaces was used to train a visual surface examination model using YOLOv3; a virtually real-time object detection algorithm. The mean average precision of the trained model was calculated to be 91%. We have conducted a series of pull-out tests to characterize the module's strength of attachments. The results of the experiments indicate that the anchoring module can withstand an applied detachment force of up to 70N and 30N when attached using teeth and vacuum suction, respectively.

show abstract

“…This arrangement enables answering the question that whether the specified region is suitable for attachment using spines; if the answer is 'no' then the vacuum suction mode is deployed. Note that, the main reason for choosing the Faster R-CNN algorithm was the ability to run this type of algorithm on NVIDIA Jetson Xavier NX with low latency at a high precision [23]. The design and implementation process of the proposed algorithm can be summarized as follows:…”

Section: Visual Examination Of Surface Properties To Make Decision On...mentioning

confidence: 99%

Faster R-CNN-based Decision Making in a Novel Adaptive Dual-Mode Robotic Anchoring System

Shahin

Sadeghian

Sareh

2021

2021 IEEE International Conference on Robotics and Automation (ICRA)

View full text Add to dashboard Cite

This paper proposes a novel adaptive anchoring module that can be integrated into robots to enhance their mobility and manipulation abilities. The module can deploy a suitable mode of attachment, via spines or vacuum suction, to different contact surfaces in response to the textural properties of the surfaces. In order to make a decision on the suitable mode of attachment, an original dataset of 100 images of outdoor and indoor surfaces was enhanced using a WGAN-GP generating an additional 200 synthetic images. The enhanced dataset was then used to train a visual surface examination model using Faster R-CNN. The addition of synthetic images increased the mean average precision of the Faster R-CNN model from 81.6% to 93.9%. We have also conducted a series of load tests to characterize the module's strength of attachments. The results of the experiments indicate that the anchoring module can withstand an applied detachment force of around 22N and 20N when attached using spines and vacuum suction on the ideal surfaces, respectively.

show abstract

Super accurate low latency object detection on a surveillance UAV

Cited by 15 publications

References 22 publications

Toward Efficient Object Detection in Aerial Images Using Extreme Scale Metric Learning

Toward Efficient Object Detection in Aerial Images Using Extreme Scale Metric Learning

Autonomous Decision Making in a Bioinspired Adaptive Robotic Anchoring Module

Faster R-CNN-based Decision Making in a Novel Adaptive Dual-Mode Robotic Anchoring System

Contact Info

Product

Resources

About