A Particle Filter Target Tracking Algorithm Combined with Efficient Outlier Detection

Abstract

The invention discloses a particle filter target tracking algorithm combined with effective abnormal point detection. On the basis of a particle filter frame, a target template is decomposed by using a PCA and an orthogonal characteristic space is established; and during follow-up tracking, a particle is projected into the characteristic space, a particle reconstruction error is calculated, and a target position is estimated by a particle probability weighted sum. For an estimated tracking result, abnormal points in the tracking result are detected by using a Lorentz estimator; and on the basis of statistics of the number of the abnormal points, whether the tracking result reaches an updated set threshold value is determined. When a certain frame number of particles are collected, the particles are projected into the original characteristic space and reconstruction errors are calculated; secondary PCA decomposition is carried out on the reconstruction errors; and a characteristic vector with a largest characteristic value among the characteristic vectors is updated to the original characteristic space. According to the invention, whether updating is needed is detected accurately and thus updating can be carried out timely, so that unnecessary updating can be avoided and tracking accuracy can be improved.

Description

technical field [0001] The invention belongs to the target tracking technology in the field of computer vision, and in particular relates to a particle filter target tracking algorithm combined with effective abnormal point detection. Background technique [0002] With the rapid development of modern technology, the application of video surveillance system has been integrated into the parking lot of many residential quarters and streets, especially in banks, airport security checks and other special occasions related to the safety of people's lives and property. Therefore, the research of video surveillance technology has become a hot spot both at home and abroad. Among them, object tracking is a technology that has been widely concerned. Although many tracking algorithms have been studied, the target tracking technology still faces many challenges, such as how to update the appearance model in time when the target appearance changes due to illumination, viewing angle, occl...

AI Technical Summary

Problems solved by technology

If the update frequency is reduced, the update will not be timely enough, resulting in wrong tracking results

Images