Three-dimensional multi-target tracking algorithm based on Poisson-multi-Bernoulli filtering

Abstract

The invention provides a three-dimensional multi-target tracking algorithm based on Poisson-multi-Bernoulli (PMBM) filtering. The main content of the algorithm comprises target formulation, target detection and target tracking. According to the process of the algorithm, first, target images are represented with parameter feature vectors; second, a deep learning network is adopted to process the target images so as to output detection results (the deep learning network is composed of a feature extractor and three pieces of parallel output header information); and last, a detection set sequenceis processed through a PMBM filter (comprising two random finite sets) in a tracking module, and an estimated sequence of a real target group is extracted and output after processing is completed. Compared with existing target tracking algorithms, the algorithm has the advantages that multiple targets can be tracked simultaneously, and tracking results are more precise; and in terms of data processing, the processing efficiency of the algorithm is higher, and associated data and even overlapping data can be processed.

Description

technical field [0001] The invention relates to the field of image processing, in particular to a three-dimensional multi-target tracking algorithm based on Poisson multi-Bernoulli filtering. Background technique [0002] With the development of computer technology and visual image processing technology, target tracking technology has gradually become a topic that scholars at home and abroad are keen to study. Target tracking technology mainly studies the detection, extraction, recognition and tracking of moving targets in video image sequences, involving machine vision, image processing and pattern recognition and other professional fields, and has broad application prospects. In human-computer interaction, object tracking technology can be used to track people's postures, actions, gestures, etc. to understand people's intentions; 3D interaction and virtual character action simulation; in medical diagnosis, target tracking technology can be used to track changes in ultraso...

AI Technical Summary

Problems solved by technology

[0004] Aiming at the problems of low data processing efficiency and inaccurate tracking results in the existing target tracking algorithm, the purpose of the present invention is to provide a three-dimensional multi-target tracking algorithm based on Poisson-Bernoulli filtering. It is represented by a parameter feature vector; then, the deep learning network is used to process the target image to output the detection result (the deep learning network is composed of a feature extractor and three parallel output header information); finally, in the tracking module, the Poisson Dobernull Use filters (including two random finite sets) to process the detection set sequence, extract and output a predicted sequence of the real target group after processing

Images