Multi-Altitude Aerial Vehicles Dataset

Custom Multi-Altitude Aerial Vehicles Dataset:

Created for publishing results for ICUAS 2023 paper "How High can you Detect? Improved accuracy and efficiency at varying altitudes for Aerial Vehicle Detection", following the abstract of the paper.

Abstract—Object detection in aerial images is a challenging task mainly because of two factors, the objects of interest being really small, e.g. people or vehicles, making them indistinguishable from the background; and the features of objects being quite different at various altitudes. Especially, when utilizing Unmanned Aerial Vehicles (UAVs) to capture footage, the need for increased altitude to capture a larger field of view is quite high. In this paper, we investigate how to find the best solution for detecting vehicles in various altitudes, while utilizing a single CNN model. The conditions for choosing the best solution are the following; higher accuracy for most of the altitudes and real-time processing ( > 20 Frames per second (FPS) ) on an Nvidia Jetson Xavier NX embedded device. We collected footage of moving vehicles from altitudes of 50-500 meters with a 50-meter interval, including a roundabout and rooftop objects as noise for high altitude challenges. Then, a YoloV7 model was trained on each dataset of each altitude along with a dataset including all the images from all the altitudes. Finally, by conducting several training and evaluation experiments and image resizes we have chosen the best method of training objects on multiple altitudes to be the mixup dataset with all the altitudes, trained on a higher image size resolution, and then performing the detection using a smaller image resize to reduce the inference performance. The main results

The creation of a custom dataset was necessary for altitude evaluation as no other datasets were available. To fulfill the requirements, the footage was captured using a small UAV hovering above a roundabout near the University of Cyprus campus, where several structures and buildings with solar panels and water tanks were visible at varying altitudes. The data were captured during a sunny day, ensuring bright and shadowless images. Images were extracted from the footage, and all data were annotated with a single class labeled as 'Car'. The dataset covered altitudes ranging from 50 to 500 meters with a 50-meter step, and all images were kept at their original high resolution of 3840x2160, presenting challenges for object detection. The data were split into 3 sets for training, validation, and testing, with the number of vehicles increasing as altitude increased, which was expected due to the larger field of view of the camera. Each folder consists of an aerial vehicle dataset captured at the corresponding altitude. For each altitude, the dataset annotations are generated in YOLO, COCO, and VOC formats. The dataset consists of the following images and detection objects:

Data
Subset
Images
Cars

50m
Train
130
269

50m
Test
32
66

50m
Valid
33
73

100m
Train
246
937

100m
Test
61
226

100m
Valid
62
250

150m
Train
244
1691

150m
Test
61
453

150m
Valid
61
426

200m
Train
246
1753

200m
Test
61
445

200m
Valid
62
424

250m
Train
245
3326

250m
Test
61
821

250m
Valid
61
823

300m
Train
246
6250

300m
Test
61
1553

300m
Valid
62
1585

350m
Train
246
10741

350m
Test
61
2591

350m
Valid
62
2687

400m
Train
245
20072

400m
Test
61
4974

400m
Valid
61
4924

450m
Train
246
31794

450m
Test
61
7887

450m
Valid
61
7880

500m
Train
270
49782

500m
Test
67
12426

500m
Valid
68
12541

mix_alt
Train
2364
126615

mix_alt
Test
587
31442

mix_alt
Valid
593
31613

It is advised to further enhance the dataset so that random augmentations are probabilistically applied to each image prior to adding it to the batch for training. Specifically, there are a number of possible transformations such as geometric (rotations, translations, horizontal axis mirroring, cropping, and zooming), as well as image manipulations (illumination changes, color shifting, blurring, sharpening, and shadowing).