Triaxial UAV Camera Tracking Function

Triaxial UAV Camera Tracking Function

Aerial cinematography has ushered in a new level of quality in video content production.triaxial uav camera tracking function However, stabilizing a drone midair and countering forces like wind so as to get precise camera shots requires highly demanding human expertise and specialized equipment. To address this, we devised an efficient UAV able to track moving objects autonomously. The system enables easy interaction with the user via a web controller and is equipped with three modes to perform various tasks.

The design consists of an onboard computer and two external cameras.triaxial uav camera tracking function The onboard computer carries out image processing to extract target information, e.g., template matching and morphological filtering. It also utilizes navigation hints, such as relative positioning and own-ship attitude estimates, to optimize the tradeoff between correct detections and false alarms in a multi-target environment. The proposed approach is robust against variations in illumination conditions, target scale and local background.

Besides visual tracking, a GPS-based type of tracker was also incorporated into this build to facilitate long-term tracking in unexpected operating environments where GPS is unreliable. The GPS tracker is a dual-mode tracking system that combines the high accuracy of a visual tracking algorithm in short-term tracking with the reliability of a GPS-based tracker for long-term tracking.

In addition to this, a 3D object model was created for the UAV that serves as a guide to efficiently navigate the drone and avoid obstacles along its path of travel. This is accomplished by leveraging proximity sensors, in this case LiDARs, to detect surrounding objects and determining the distance between the target object and the drone. This information is then used to compute an optimal trajectory.

The DB-Tracker tracker outperformed other prominent trackers on the VisDrone MOT and UAVDT datasets. Its superiority stems from its effective fusion of position and appearance information, which results in improved performance on all evaluation metrics. For example, the DB-Tracker outperforms Deep SORT in terms of MOTA and MOTP. Moreover, it performs better than ByteTrack in terms of similarity tracking and background noise filtering. It also outperforms Deep OC-SORT in terms of camera motion compensation and ID feature association.

In addition, the DB-Tracker can track multiple objects simultaneously in both still and moving scenes. This is possible because it utilizes a more sophisticated loss function, which accounts for both the location and orientation of each object. In contrast, other trackers use a single-loss function that only accounts for the location of each object. Furthermore, it is worth noting that the DB-Tracker outperforms all other trackers in terms of MOTA and MOTP when using the VisDrone MOT test set. For this reason, it can be considered as a state-of-the-art tracker in UAV aerial video applications. In the future, we plan to improve the tracker by further integrating intelligent algorithms to interpret the fused sensor data more effectively. For example, we plan to employ neural networks for both the image and time-series input data. In this way, they can help to identify patterns and anomalies more effectively than conventional methods.

Tags:camera drone camera | drone camera