Using YOLO-based pedestrian detection for monitoring UAV

Zhang, Depei; Shao, Yanhua; Mei, Yanying; Chu, Hsin‐Sen; Zhang, Xiaoqiang; Zhan, Huayi; Rao, Yunbo

doi:10.1117/12.2524219

Cited by 6 publications

(7 citation statements)

References 5 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The full implementation would mount the Jetson on a drone with a dedicated power source and cellular modem to communicate out results; these are both relatively inexpensive pieces of commodity hardware. Zhang et al (2019) have already demonstrated the feasibility of attaching the Jetson to a drone for use in real-time inference.…”

Section: Discussionmentioning

confidence: 99%

“…We did, however, run model inference on the hold-out test video on our edge device. The device can be mounted on the drone and attached to a power source ( Zhang et al, 2019 ).…”

Section: Methodsmentioning

confidence: 99%

“…Running object detection on-board the drone reduces the amount of time needed for a human to review footage, and can also give game managers real-time geo-locations of animals and poachers. Use of UAVs with embedded hardware in pedestrian detection ( Zhang et al, 2019 ) and tracking ( Smedt De, Hulens & Goedeme, 2015 ) paved the way for other related applications such as animal detection. Until recently, animal real-time object detection and alert had not been fully automated on-board drones due to limited processing power, and limited internet connection in the field.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Protecting endangered megafauna through AI analysis of drone images in a low-connectivity setting: a case study from Namibia

Hua

Martin

Shen

et al. 2022

PeerJ

View full text Add to dashboard Cite

Assessing the numbers and distribution of at-risk megafauna such as the black rhino (Diceros bicornis) is key to effective conservation, yet such data are difficult to obtain. Many current monitoring technologies are invasive to the target animals and expensive. Satellite monitoring is emerging as a potential tool for very large animals (e.g., elephant) but detecting smaller species requires higher resolution imaging. Drones can deliver the required resolution and speed of monitoring, but challenges remain in delivering automated monitoring systems where internet connectivity is unreliable or absent. This study describes a model built to run on a drone to identify in situ images of megafauna. Compared with previously reported studies, this automated detection framework has a lower hardware cost and can function with a reduced internet bandwidth requirement for local network communication. It proposes the use of a Jetson Xavier NX, onboard a Parrot Anafi drone, connected to the internet throughout the flight to deliver a lightweight web-based notification system upon detection of the target species. The GPS location with the detected target species images is sent using MQ Telemetry Transport (MQTT), a lightweight messaging protocol using a publisher/subscriber architecture for IoT devices. It provides reliable message delivery when internet connection is sporadic. We used a YOLOv5l6 object detection architecture trained to identify a bounding box for one of five objects of interest in a frame of video. At an intersection over union (IoU) threshold of 0.5, our model achieved an average precision (AP) of 0.81 for black rhino (our primary target) and 0.83 for giraffe (Giraffa giraffa). The model was less successful at identifying the other smaller objects which were not our primary targets: 0.34, 0.25, and 0.42 for ostrich (Struthio camelus australis), springbok (Antidorcas marsupialis) and human respectively. We used several techniques to optimize performance and overcome the inherent challenge of small objects (animals) in the data. Although our primary focus for the development of the model was rhino, we included other species classes to emulate field conditions where many animal species are encountered, and thus reduce the false positive occurrence rate for rhino detections. To constrain model overfitting, we trained the model on a dataset with varied terrain, angle and lighting conditions and used data augmentation techniques (i.e., GANs). We used image tiling and a relatively larger (i.e., higher resolution) image input size to compensate for the difficulty faced in detecting small objects when using YOLO. In this study, we demonstrated the potential of a drone-based AI pipeline model to automate the detection of free-ranging megafauna detection in a remote setting and create alerts to a wildlife manager in a relatively poorly connected field environment.

show abstract

Section: Discussionmentioning

confidence: 99%

“…We did, however, run model inference on the hold-out test video on our edge device. The device can be mounted on the drone and attached to a power source ( Zhang et al, 2019 ).…”

Section: Methodsmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Protecting endangered megafauna through AI analysis of drone images in a low-connectivity setting: a case study from Namibia

Hua

Martin

Shen

et al. 2022

PeerJ

View full text Add to dashboard Cite

show abstract

“…In [1,19], YOLOv3 and YOLOv4 algorithms were adopted to detect helmet wearing and worker falls, respectively. In addition to personnel detection, the YOLO detector was combined with UAV to perform real-time detection of the designated area [20,21], overcoming the problem of target omission and incorrect detection caused by pedestrian density and pedestrian occlusion under a fixed viewing angle.…”

Section: One-stage Detection Methodsmentioning

confidence: 99%

CaSnLi-YOLO: construction site multi-target detection method based on improved YOLOv5s

Hao,

Zhang,

et al. 2024

Meas. Sci. Technol.

View full text Add to dashboard Cite

To address the problem of the frequent occurrence of major casualties during construction, a lightweight multi-target detection model based on YOLOv5s, named CaSnLi-YOLO, was proposed for the detection of multiple targets consisting of construction workers and various mechanical equipment at construction sites. In this study, the introduction of the Coordinate Attention (CA) mechanism along with DWConv and C3Ghost based on GhostNet, enhanced the expression and detection accuracy of the model, effectively reducing the number of parameters. Considering the environmental characteristics of construction sites, a detection box filtering strategy based on Soft Non-Maximum Suppression (Soft-NMS) was employed, to effectively alleviate the issue of missed detections of occluded targets. Experimental results demonstrate the significant superiority of the proposed CaSnLi-YOLO over current mainstream detection models, such as faster region-based convolutional neural network (Faster R-CNN) and single-shot detector (SSD). The proposed CaSnLi-YOLO has a parameter number of 5.96×106, which is 15.2% less than that of the original YOLOv5s model, further improving precision, recall rate, mAP@0.5, and mAP@0.5:0.95 by 0.6%, 0.6 %, 0.2%, and 2.3%, respectively. The improved YOLOv5s model proposed in this study achieved significant improvements in multi-target detection at construction sites. The model demonstrated effective enhancements while significantly reducing parameter count and is expected to be deployed in small-edge devices for real-time security monitoring at construction sites.

show abstract

“…In previous work on pedestrian detection, Zhang et al [14] showed the effectiveness of gradient boosting. Such a technique can also be applied to action detection and recognition from single images in UAV data.…”

Section: Introductionmentioning

confidence: 99%

Detecting Human Actions in Drone Images Using YoloV5 and Stochastic Gradient Boosting

Ahmad

Cavazza

Matsuo

et al. 2022

Sensors

View full text Add to dashboard Cite

Human action recognition and detection from unmanned aerial vehicles (UAVs), or drones, has emerged as a popular technical challenge in recent years, since it is related to many use case scenarios from environmental monitoring to search and rescue. It faces a number of difficulties mainly due to image acquisition and contents, and processing constraints. Since drones’ flying conditions constrain image acquisition, human subjects may appear in images at variable scales, orientations, and occlusion, which makes action recognition more difficult. We explore low-resource methods for ML (machine learning)-based action recognition using a previously collected real-world dataset (the “Okutama-Action” dataset). This dataset contains representative situations for action recognition, yet is controlled for image acquisition parameters such as camera angle or flight altitude. We investigate a combination of object recognition and classifier techniques to support single-image action identification. Our architecture integrates YoloV5 with a gradient boosting classifier; the rationale is to use a scalable and efficient object recognition system coupled with a classifier that is able to incorporate samples of variable difficulty. In an ablation study, we test different architectures of YoloV5 and evaluate the performance of our method on Okutama-Action dataset. Our approach outperformed previous architectures applied to the Okutama dataset, which differed by their object identification and classification pipeline: we hypothesize that this is a consequence of both YoloV5 performance and the overall adequacy of our pipeline to the specificities of the Okutama dataset in terms of bias–variance tradeoff.

show abstract

Using YOLO-based pedestrian detection for monitoring UAV

Cited by 6 publications

References 5 publications

Protecting endangered megafauna through AI analysis of drone images in a low-connectivity setting: a case study from Namibia

Protecting endangered megafauna through AI analysis of drone images in a low-connectivity setting: a case study from Namibia

CaSnLi-YOLO: construction site multi-target detection method based on improved YOLOv5s

Detecting Human Actions in Drone Images Using YoloV5 and Stochastic Gradient Boosting

Contact Info

Product

Resources

About