Views: 0 Author: Site Editor Publish Time: 2025-01-10 Origin: Site
On January 9, 2025, according to GlobeSpec, drone tracking has become increasingly important in both civil and military domains. Whether it's the recent frequent sightings of mysterious unidentified aerial phenomena (UAP) on the US East Coast or drone warfare expanding to distant cities and defense positions on the other side of the world, these events highlight the significant challenge of monitoring drones. While many technologies are already available to achieve this goal, most suffer from slow updates or inaccurate detection, both of which struggle to keep up with fast-moving small drones. A team from the Vienna University of Technology believes they may have found a way to eliminate the weaknesses of previous drone tracking technologies, and they have demonstrated this by using telescopes.
Current drone tracking technologies can generally be divided into two categories: detectors and trackers. While there are many multispectral methods to detect drones using radar, acoustics, or other means, optical tracking remains the most widely used method because it is the easiest to improve through non-expert feedback. Traditional detection methods often have a high false positive rate, making accurate identification of drones difficult. In recent years, deep learning algorithms have made progress in improving drone detection accuracy. However, although these algorithms are more accurate than earlier classifiers and filters, their processing speeds are still very slow. Some drones now move at speeds of up to 20 meters per second, far outpacing the processing speed of these detection algorithms. By the time a drone is detected, it may have already vanished without a trace. Tracking algorithms usually run faster, but their accuracy is not as high as detectors. Tracking algorithms plot trajectories and speeds by comparing image frames of the target, which makes them highly susceptible to interference from external objects, obstructions, weather, or misclassifications. Even if the drone is still within the imaging system’s field of view, the tracking model may easily get confused and lose track of the drone.
The team from the Vienna University of Technology believes they have found a way to combine the accuracy of detectors with the speed of trackers to create a system that is both faster and more accurate than either system alone. Their system simultaneously records image frames from two different memory sources for analysis. Different central processing units (CPUs) run the detection and tracking algorithms separately. Because the detection algorithm is slower, it processes only 20% of the captured image frames, while the tracking algorithm processes each frame sequentially. The tracking model is corrected based on the output from the detector. The system relies on a reliability metric to decide when to start or end tracking a target. The deep learning algorithm provides a confidence score for each detection, while the tracker supports providing the likelihood of correctly tracking and estimating the target’s position. The higher the tracker’s reliability, the smarter the collaboration between the detector and the tracker. When the confidence from the deep learning-based detector exceeds the reliability of the tracker, the system can reinitialize the tracker to ensure the most reliable information is used. Additionally, the tracker’s reliability enables the system to dynamically adapt to changes in the tracking environment. For example, if lighting conditions or drone behavior causes the tracker’s reliability to decrease, the system can adjust by relying more on the detector or reinitializing the tracker as needed.
An important consideration when tracking drones is how to handle new tracking information. Since the field of view of a telescope is relatively small, a fast-moving drone can quickly move out of view. Therefore, the team built a control system that can move the telescope to track the drone. The calculated trajectory is fed into the telescope’s pan-tilt controller. To validate their system, the researchers conducted field tests, tracking a series of drones against both clear (sky) and complex (such as rows of buildings or trees) backgrounds. They then compared the results with those from independent detector systems and tracker systems. In tests with a clear background, their system performed similarly to existing methods, but in complex backgrounds, it significantly outperformed algorithms based on independent detectors (accuracy improved by 6%) or independent trackers (accuracy improved by 14%). Moreover, the system has another advantage that neither of the individual systems can match. Estimating the flight path is key to tracking, which is not simply about detecting the presence of a drone in the image. To accurately estimate the flight path, a dedicated algorithm for flight path calculation must obtain two pieces of location information from the detection algorithm—i.e., the precise position of the drone in each frame. In this regard, the performance of the combined detector/tracker method was 49% higher than using the detector alone. While using only the tracker produced better results, the flight path estimated by it did not necessarily guarantee accuracy.
Impressively, even when the drone flew out of the detector’s view and the distance between the two reached 4 kilometers, the method still worked. Even with incremental improvements in drone detection, this could save lives in certain situations, such as on the battlefield in Ukraine. However, the system is not yet ready for large-scale deployment. It still requires a large computer equipped with a graphics processing unit (GPU) to run the full algorithm, and future work will include integrating an additional telescope with a larger field of view to further improve detection accuracy and tracking capabilities. Achieving accurate drone tracking in all scenarios remains a difficult challenge. Nevertheless, given the interest and incentives from military and law enforcement sectors, this technology is expected to see further improvements in the near future.
