Multiple Object Tracking (MOT) is a classical task in the field of computer vision, which aims to identify and track all objects in a video scene and assign a unique ID number to each object. Tracking-by-Detection (TBD) paradigm has become the mainstream framework for MOT due to its high Tracking accuracy. With the development of UAV technology, MOT research for UAV video has important military and civilian value. However, it faces challenges such as class imbalance, many small targets, and occlusion of targets in the scene, which makes it difficult to correctly match and continuously track the targets. We propose a new algorithm for the MOT problem in the UAV scenario. On the one hand, to solve the class imbalance problem of small targets, a dynamic adjustment parameter adjustment method based on the gradient information of training samples is proposed to improve the generalization ability of traditional loss function in multi-class target tracking. On the other hand, to improve the accuracy of inter-frame matching, this paper introduces a new feature similarity calculation method, which is based on the Wasserstein distance and optimizes the matching process according to the weight allocation mechanism of feature importance. Finally, the effectiveness of the proposed algorithm is verified on the VisDroneMOT2019 dataset. The results show that compared with the existing MOT algorithm, the proposed algorithm has significant improvements in tracking accuracy, trajectory integrity and identity maintenance, achieving 38.8% MOTA and 52.8% IDF1, which are better than the existing state-of-the-art tracking algorithms.
With the improvement of hardware computing power, the application of deep learning methods in the field of remote sensing is increasing. This paper summarizes the progress of deep learning methods in remote sensing image object detection in recent years. The main methods of deep learning methods to extract and use target feature information in various target detection tasks are summarized. Finally, the application trend of deep learning methods in the field of remote sensing image detection is prospected.
Target detection of arbitrary shape is widely used in remote sensing image processing, the case segmentation method based on contour regression is very similar to the target detection method. At present, many scholars have studied case segmentation using deep learning framework, case segmentation based on contour regression can be regarded as a simple extension of object detection, it is a more accurate target detection. In this paper, the contour regression method based on Fourier operator is used for accurate target detection, the Fourier feature vectors in a group of frequencies are predicted at each position of the feature map, and then the target contour point sequence is reconstructed in the image space domain through inverse Fourier transform, The simulation results show that the detection accuracy of the proposed method is more than 10% higher than that of the classical method.
The theoretical model of atmospheric refractive index based on the standard atmospheric environment cannot explain the atmospheric refractive index when optical satellites detect targets. Aiming at this problem, a method for estimating the refractive index of the atmosphere based on multispectral stellar observation data is proposed. Based on optical satellites’ multispectral stellar observation data, according to the principle of deflection when the stellar light passes through the atmosphere, the optical path model of the stellar light refracted by the atmosphere is established under the assumption of a layered spherical atmosphere. A method of using multispectral segment stars to measure the actual light and the iterative forward feedback of stellar theoretical light is proposed, to estimate the refractive index of each layer of the atmosphere of the layered atmosphere to different spectral segments of light.
A problem of state estimation with a new constraints named incomplete nonlinear constraint is considered. The targets are often move in the curve road, if the width of road is neglected, the road can be considered as the constraint, and the position of sensors, e.g., radar, is known in advance, this info can be used to enhance the performance of the tracking filter. The problem of how to incorporate the priori knowledge is considered. In this paper, a second-order sate constraint is considered. A fitting algorithm of ellipse is adopted to incorporate the priori knowledge by estimating the radius of the trajectory. The fitting problem is transformed to the nonlinear estimation problem. The estimated ellipse function is used to approximate the nonlinear constraint. Then, the typical nonlinear constraint methods proposed in recent works can be used to constrain the target state. Monte-Carlo simulation results are presented to illustrate the effectiveness proposed method in state estimation with incomplete constraint.
A motion model for the hypersonic boost-glide aircraft(HBG) was proposed in this paper, which also analyzed the precision of model through simulation. Firstly the trajectory of HBG was analyzed, and a scheme which divide the trajectory into two parts then build the motion model on each part. Secondly a restrained model of boosting stage and a restrained model of J2 perturbation were established, and set up the observe model. Finally the analysis of simulation results show the feasible and high-accuracy of the model, and raise a expectation for intensive research.
In space based optical surveillance systems, the targets could be points, spots or special shape, it depends on the sensors’ parameters. This paper proposed a method to divide the target’s surface into infinitesimal according to a simplified model, then the ray tracing method is used. Typical effects resulting from sensors’ parameters are considered. Imaging simulation results indicates that target’s plume flow images produced by this method are logical and can be applied to validate data processing algorithms.
In order to validate the performance of data processing algorithms cheaply and efficiently, imaging simulation is significant. This pa per discusses an imaging simulation method for space-based optical sensors which have wide field of vision and scanning mechanisms. To avoid background’s saltation and to make simulation real, the proposed method uses the data from Moderate-resolution Imaging Spectroradiometer (MODIS) as original input. Pre-treatments for MODIS data including periodic extension are applied to enlarge the background. Then scan-line sensor is modelled and ray tracing method is used. Typical effects resulting from the sun and detectors are considered. Finally, imaging simulation results indicates that images produced by this method are logical and can be applied to validate data processing algorithms.
ABSTRACT In Space-based optical system, during the tracking for closely spaced objects (CSOs), the traditional method with a constant false alarm rate(CFAR) detecting brings either more clutter measurements or the loss of target information. CSOs can be tracked as Extended targets because their features on optical sensor’s pixel-plane. A pixel partition method under the framework of Markov random field(MRF) is proposed, simulation results indicate: the method proposed provides higher pixel partition performance than traditional method, especially when the signal-noise-rate is poor.
In order to surveillance the geostationary (GEO) objects, including man-made satellites and space debris, more efficiently, a space-based optical surveillance system was designed in this paper. A strategy to observe the pinch point region was selected because of the GEO objects’ dynamics features. That strategy affects the surveillance satellites orbital type and sensor pointing strategy. In order to minimize total surveillance satellites and the revisit time for GEO objects, a equation was set. More than 700 GEO objects’ TLE from NASA’s website are used for simulation. Results indicate that the revisit time of the surveillance system designed in this paper is less than 24 hours, more than 95% GEO objects can be observed by the designed system.
Access to the requested content is limited to institutions that have purchased or subscribe to SPIE eBooks.
You are receiving this notice because your organization may not have SPIE eBooks access.*
*Shibboleth/Open Athens users─please
sign in
to access your institution's subscriptions.
To obtain this item, you may purchase the complete book in print or electronic format on
SPIE.org.
INSTITUTIONAL Select your institution to access the SPIE Digital Library.
PERSONAL Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.