The Aerial Elephant Dataset: A New Public Benchmark for Aerial Object Detection.

Johannes Naude, Deon Joubert; Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2019, pp. 48-55

Abstract


Aerial surveying is a key tool for effective wildlife management. However, the high costs associated with large scale surveys means that this tool is often underutilized. We believe that computer vision can be used to dramatically decrease the costs associated with surveying, while at the same time improving the consistency of results. We present the Aerial Elephant Dataset, a challenging dataset to enable research on game detection under real-world conditions. The dataset consists of 2101 images containing a total of 15 511 African bush elephants in their natural habitats, imaged with a consistent methodology over a range of background types, resolutions and times-of-day. A baseline algorithm for elephant detection is trained and tested to demonstrate the feasibility of the proposed task. The algorithm is used in a larger system, where false positive rejection and counting of densely spaced individuals is aided by a human-in-the-loop. We evaluate the performance of this system against traditional methods by performing surveys in tandem with professional human surveying crews and comparing results in terms of detections missed, man-hours spent and cost.

Related Material


[pdf]
[bibtex]
@InProceedings{Naude_2019_CVPR_Workshops,
author = {Naude, Johannes and Joubert, Deon},
title = {The Aerial Elephant Dataset: A New Public Benchmark for Aerial Object Detection.},
booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops},
month = {June},
year = {2019}
}