Improved KCF target tracking method combined with lightweight SSD

Abstract

The invention discloses an improved KCF target tracking method combined with a lightweight SSD. The method comprises the steps: carrying out the feature extraction of an initial frame target through a lightweight SSD target detection frame, enabling the extracted features to pass through a prediction module, a non-maximum suppression module and an output detection module in sequence, and transmitting the initial frame target information to a KCF for initialization; constructing a cyclic matrix for a to-be-tracked target by the KCF, adopting ridge regression to train a position filter, and adopting a binary tree scale search strategy to select an optimal scale near a position determined by the position filter for a scale problem; for the shielding problem occurring in target tracking, selecting average peak energy as a judgment index, detecting whether the target is shielded or not, and determining whether template updating is carried out or not according to a judgment result. According to the method, the problem of low efficiency caused by manual marking of the tracking target in KCF target tracking is effectively solved, and the precision and accuracy of target tracking are improved under the conditions of scale change, shielding and the like of the target.

Description

technical field [0001] The invention belongs to the technical field of visual target tracking, in particular to an improved KCF target tracking method combined with a lightweight SSD. technical background [0002] The detection and tracking of moving targets is an important research content in the field of computer vision research, and has important applications in civil and military fields such as intelligent monitoring, military reconnaissance, and human-computer interaction. The correlation filtering (Correlation Filtering, CF) tracking algorithm is a typical discriminative method. In 2010, Bolem introduced the correlation in the field of signal processing into the field of computer vision, and proposed the minimum square error output and MOSSE filter, which achieved amazing tracking speed. 615FPS. On this basis, the CSK algorithm extends dense sampling, and uses kernel techniques to realize high-dimensional operations in low-dimensional spaces, avoiding the curse of dim...

AI Technical Summary

Problems solved by technology

Although the target tracking algorithm based on kernel correlation filter performs well in terms of speed, it is not satisfactory when it is directly used to process the video sequence shot from the perspective of the UAV, especially when the pictures taken by the UAV have target scale changes and deformations. Complex situations such as , occlusion, etc., often lead to the tracking failure of traditional kernel correlation filtering algorithms

Martin proposed the DSST algorithm, which introduces a scale filter based on the two-dimensional position filter, collects 33 scales at the center of the target for similarity comparison, and selects the maximum response as the best matching scale for the target. However, due to the violent matching method, the computational complexity of the algorithm is relatively high; secondly, since the initial frame target is determined by manual labeling, the real-time performance of the algorithm operation is reduced

Images