UASs (Unmanned Aerial Systems) should have the ability to identify various objects located in the immediate environment and track their movements a time passes to navigate their surrounding environment and accomplish tasks. However, researchers have failed to equip unmanned aerial vehicles with multi-object tracking, which is a highly challenging task today.
Researchers have recently developed a novel deep learning technique that would enable UASs to track numerous objects in their environment. The technique is a significant contribution to the Unmanned Aerial Vehicle System Market. It has the potential to lead to the creation of more responsive and high-performing autonomous flying systems. The team presented a powerful object tracking architecture to adapt to the noise associated with real-time situations. Further, they also brought forth a kinematic prediction model referred to as a Deep EKF (Deep Extended Kalman Filter). They used a sequence-to-sequence architecture to foresee entity trajectories within latent space.
A KF (Kalman Filter) refers to an algorithm that can estimate unknown variables once it is fed measurements of different periods. The researchers' multi-object tracking approach in the present study is an advanced version of KF accomplished through deep learning techniques.
The team trained their system to use a computational attention mechanism and an acquired image embedding to understand the necessity of various parts of an image towards envisaging changes and future states. By evaluating images with a convolutional neural network (CNN) encoder pre-trained with Siamese neural networks, the model uses similarity measures to calculate distances between objects.
The deep learning technique was also tested through annotated video footage acquired with the help of a camera attached to a fixed-wing UAS. The labelled video sequenced received comprised of several moving objects involving people and vehicles. The preliminary evaluations demonstrated that DeepEKF architecture performed remarkably and produced outstanding results, even outperforming standard KF algorithms used for multi-object tracking. In the near future, once the system is appropriately researched upon and developed further, it could surely help enhance the abilities of different types of UASs.
Researchers now plan to add a more wide-ranging evaluation for the system to achieve long-term tracking (re-identification) component. They will also investigate the combination of visual and kinematic scores for the similarity fuser components provided to the environment and track states.
Related Reports:
Global Vehicle Intelligence Systems Market Research Report - Industry Analysis, Size, Share, Growth, Trends and Forecast Till 2026
Global Biometric Vehicle Access System Market 2021 by Manufacturers, Regions, Type and Application, Forecast to 2026
Global Vehicle Security System Market 2019 by Company, Regions, Type and Application, Forecast to 2024
Global Vehicle Wash Systems Market 2019 by Manufacturers, Regions, Type and Application, Forecast to 2024
