TY  - JOUR
A1  - Stiemer, Luc Nicolas
A1  - Thoma, Andreas
A1  - Braun, Carsten
T1  - MBT3D: Deep learning based multi-object tracker for bumblebee 3D flight path estimation
JF  - PLoS ONE
N2  - This work presents the Multi-Bees-Tracker (MBT3D) algorithm, a Python framework implementing a deep association tracker for Tracking-By-Detection, to address the challenging task of tracking flight paths of bumblebees in a social group. While tracking algorithms for bumblebees exist, they often come with intensive restrictions, such as the need for sufficient lighting, high contrast between the animal and background, absence of occlusion, significant user input, etc. Tracking flight paths of bumblebees in a social group is challenging. They suddenly adjust movements and change their appearance during different wing beat states while exhibiting significant similarities in their individual appearance. The MBT3D tracker, developed in this research, is an adaptation of an existing ant tracking algorithm for bumblebee tracking. It incorporates an offline trained appearance descriptor along with a Kalman Filter for appearance and motion matching. Different detector architectures for upstream detections (You Only Look Once (YOLOv5), Faster Region Proposal Convolutional Neural Network (Faster R-CNN), and RetinaNet) are investigated in a comparative study to optimize performance. The detection models were trained on a dataset containing 11359 labeled bumblebee images. YOLOv5 reaches an Average Precision of AP = 53, 8%, Faster R-CNN achieves AP = 45, 3% and RetinaNet AP = 38, 4% on the bumblebee validation dataset, which consists of 1323 labeled bumblebee images. The tracker’s appearance model is trained on 144 samples. The tracker (with Faster R-CNN detections) reaches a Multiple Object Tracking Accuracy MOTA = 93, 5% and a Multiple Object Tracking Precision MOTP = 75, 6% on a validation dataset containing 2000 images, competing with state-of-the-art computer vision methods. The framework allows reliable tracking of different bumblebees in the same video stream with rarely occurring identity switches (IDS). MBT3D has much lower IDS than other commonly used algorithms, with one of the lowest false positive rates, competing with state-of-the-art animal tracking algorithms. The developed framework reconstructs the 3-dimensional (3D) flight paths of the bumblebees by triangulation. It also handles and compares two alternative stereo camera pairs if desired.
Y1  - 2023
U6  - http://dx.doi.org/10.1371/journal.pone.0291415
SN  - 1932-6203
N1  - Corresponding author: Luc Nicolas Stiemer
VL  - 18
IS  - 9
PB  - PLOS
CY  - San Fancisco
ER  - 
TY  - JOUR
A1  - Thoma, Andreas
A1  - Thomessen, Karolin
A1  - Gardi, Alessandro
A1  - Fisher, A.
A1  - Braun, Carsten
T1  - Prioritising paths: An improved cost function for local path planning for UAV in medical applications
JF  - The Aeronautical Journal
N2  - Even the shortest flight through unknown, cluttered environments requires reliable local path planning algorithms to avoid unforeseen obstacles. The algorithm must evaluate alternative flight paths and identify the best path if an obstacle blocks its way. Commonly, weighted sums are used here. This work shows that weighted Chebyshev distances and factorial achievement scalarising functions are suitable alternatives to weighted sums if combined with the 3DVFH* local path planning algorithm. Both methods considerably reduce the failure probability of simulated flights in various environments. The standard 3DVFH* uses a weighted sum and has a failure probability of 50% in the test environments. A factorial achievement scalarising function, which minimises the worst combination of two out of four objective functions, reaches a failure probability of 26%; A weighted Chebyshev distance, which optimises the worst objective, has a failure probability of 30%. These results show promise for further enhancements and to support broader applicability.
KW  - Path planning
KW  - Cost function
KW  - Multi-objective optimization
Y1  - 2023
U6  - http://dx.doi.org/10.1017/aer.2023.68
SN  - 0001-9240 (Print)
SN  - 2059-6464 (Online)
IS  - First View
SP  - 1
EP  - 18
PB  - Cambridge University Press
CY  - Cambridge
ER  - 
TY  - CHAP
A1  - Thoma, Andreas
A1  - Stiemer, Luc
A1  - Braun, Carsten
A1  - Fisher, Alex
A1  - Gardi, Alessandro G.
T1  - Potential of hybrid neural network local path planner for small UAV in urban environments
T2  - AIAA SCITECH 2023 Forum
N2  - This work proposes a hybrid algorithm combining an Artificial Neural Network (ANN) with a conventional local path planner to navigate UAVs efficiently in various unknown urban environments. The proposed method of a Hybrid Artificial Neural Network Avoidance System is called HANNAS. The ANN analyses a video stream and classifies the current environment. This information about the current Environment is used to set several control parameters of a conventional local path planner, the 3DVFH*. The local path planner then plans the path toward a specific goal point based on distance data from a depth camera. We trained and tested a state-of-the-art image segmentation algorithm, PP-LiteSeg. The proposed HANNAS method reaches a failure probability of 17%, which is less than half the failure probability of the baseline and around half the failure probability of an improved, bio-inspired version of the 3DVFH*. The proposed HANNAS method does not show any disadvantages regarding flight time or flight distance.
Y1  - 2023
U6  - http://dx.doi.org/10.2514/6.2023-2359
N1  - AIAA SCITECH 2023 Forum, 23-27 January 2023, National Harbor, MD & Online
PB  - AIAA
ER  -