A multi-target tracking method for space-based surveillance

Abstract

The invention discloses a multi-target tracking method oriented to space-based surveillance, which belongs to the field of aerial surveillance. Firstly, the position coordinate information of all suspected targets in the scene graph to be detected is obtained. A suspected target is extracted and fused using a feature extraction network based on appearance information and a feature extraction network based on motion information. Similarly, for all suspected targets in each frame of image, the feature vectors that combine appearance and motion information are calculated separately. Each suspected target in the previous frame and the current frame is traversed in pairs to obtain the similarity between each two suspected targets, and the suspected target is used as a point, and the similarity is used as the weight on the edge to construct a graph convolutional neural network. For each point in the current frame, the feature vector of another point on the connected edge is propagated to the point through the weight and updated, and the feature similarity of each two suspected targets is recalculated to ensure the one-to-one matching of multiple targets between two frames. The invention can efficiently propagate target feature vectors, and is easy to construct and train.

Description

technical field [0001] The invention belongs to the field of aviation surveillance, in particular to a multi-target tracking method for space-based surveillance. Background technique [0002] In recent years, with the continuous opening up of the country's use authority in the low-altitude field, various technologies and tasks in the space-based surveillance platform have developed rapidly; such as auxiliary security management, special scene surveillance, etc. Safety monitoring, security management, intelligence collection and analysis, tracking and detection of abnormal pedestrians or foreign objects on rail tracks, etc. are applied to rail monitoring tasks. UAVs can systematically monitor abnormal pedestrians or foreign objects on rails from an air-based perspective to ensure railway transportation safety and efficiency. Among these tasks, the statistics and analysis of multi-target tracking in complex scenes is one of the most important links. In many scenes, especially...

AI Technical Summary

Problems solved by technology

However, this kind of method cannot guarantee the one-to-one matching of targets between two frames, and does not allow the information propagation of all matching results in two frames, and thus cannot achieve the optimal matching solution for multi-target tracking, which limits its effect to a certain extent.

Images