In many applications of airborne visual techniques for unmanned aerial vehicles (UAVs), lightweight sensors and efficient visual positioning and tracking algorithms are essential in a GNSS-denied environment. Meanwhile, many tasks require the ability of recognition, localization, avoiding, or flying pass through these dynamic obstacles. In this paper, for a small UAV equipped with a lightweight monocular sensor, a single-frame parallel-features positioning method (SPPM) is proposed and verified for a real-time dynamic target tracking and ingressing problem. The solution is featured with systematic modeling of the geometric characteristics of moving targets, and the introduction of numeric iteration algorithms to estimate the geometric center of moving targets. The geometric constraint relationships of the target feature points are modeled as non-linear equations for scale estimation. Experiments show that the root mean square error percentage of static target tracking is less than 1.03% and the root mean square error of dynamic target tracking is less than 7.92 cm. Comprehensive indoor flight experiments are conducted to show the real-time convergence of the algorithm, the effectiveness of the solution in locating and tracking a moving target, and the excellent robustness to measurement noises.

In this study, a set of benchmarks for object tracking with motion parameters (OTMP) was first designed. The sample images were matched with the spatial depth of the camera, the pose of the camera, and other spatial parameters for the training of the detection model. Then, a Fast Depth-Assisted Single-Shot MultiBox Detector (FDA-SSD) algorithm suitable for 3D target tracking was proposed by combining the depth information of the sample into the original Single-Shot MultiBox Detector (SSD). Finally, an FDA-SSD-based monocular motion platform target detection and tracking algorithm framework were established. Specifically, the spatial geometric constraints of the target were adapted to solve the target depth information, which was fed back to the detection model. Then, the normalized depth information of the target was employed to select the feature window of the convolutional layer for the detector at a specific scale. This significantly reduces the computational power for simultaneously calculating detectors of all scales. This framework effectively combines the two-dimensional detection model and the three-dimensional positioning algorithm. Compared with the original SSD method, the network model designed in this study has fewer actual operating parameters; the measured detection operation speed was increased by about 18.1% on average; the recognition rate was maintained at a high level consistent with that of the original SSD. Furthermore, several groups of experiments were conducted on target detection and target space tracking based on monocular motion platforms indoors. The root mean square error (RMSE) of the spatial tracking trajectory was less than 4.72 cm. The experimental results verified that the algorithm framework in this study can effectively realize tasks such as visual detection, classification, and spatial tracking based on a monocular motion platform.

This paper proposes an image augmentation model of limited samples on the mobile platform for object tracking. The augmentation method mainly aims at the detection failure caused by the small number of effective samples, jitter of tracking platform, and relative rotation between camera and object in the tracking process. Aiming at the object tracking problem, we first propose to use geometric projection transformation, multi-directional overlay blurring, and random background filling to improve the generalization ability of samples. Then, selecting suitable traditional augmentation methods as the supplements, an image augmentation model with an adjustable probability factor is provided to simulate various kinds of samples to help the detection model carry out more reliable training. Finally, combined with a spatial localization algorithm based on geometric constraints proposed by the author’s previous work, a framework for object tracking with an image augmentation method is proposed. SSD, YOLOv3, YOLOv4, and YOLOx are adopted in the experiment of this paper as the detection models. And a large number of object recognition and object tracking experiments are carried out by combining with common data sets OTB50 and OTB100 as well as the OTMP data set proposed by us for mobile platform. The augmented module proposed in this paper is conducive for the detection model to improve the detection accuracy by at least 10%. Especially for objects with planar characteristics, the affine and projection transformation used in this paper can greatly improve the detection accuracy of the model. Based on the object tracking framework of our augmented model, the RMSE is estimated to be less than 4.21 cm in terms of the actual tracking of indoor objects.

In this study, a multi-level scale stabilizer intended for visual odometry (MLSS-VO) combined with a self-supervised feature matching method is proposed to address the scale uncertainty and scale drift encountered in the field of monocular visual odometry. Firstly, the architecture of an instance-level recognition model is adopted to propose a feature matching model based on a Siamese neural network. Combined with the traditional approach to feature point extraction, the feature baselines on different levels are extracted, and then treated as a reference for estimating the motion scale of the camera. On this basis, the size of the target in the tracking task is taken as the top-level feature baseline, while the motion matrix parameters as obtained by the original visual odometry of the feature point method are used to solve the real motion scale of the current frame. The multi-level feature baselines are solved to update the motion scale while reducing the scale drift. Finally, the spatial target localization algorithm and the MLSS-VO are applied to propose a framework intended for the tracking of target on the mobile platform. According to the experimental results, the root mean square error (RMSE) of localization is less than 3.87 cm, and the RMSE of target tracking is less than 4.97 cm, which demonstrates that the MLSS-VO method based on the target tracking scene is effective in resolving scale uncertainty and restricting scale drift, so as to ensure the spatial positioning and tracking of the target.