Group target trajectory analysis model based on overall graph averaging model

Abstract

The invention discloses a group target trajectory analysis model based on an overall graph averaging model, and the model is characterized in that the input is a trajectory segment of a target, and the output is a complete motion trajectory of the target under multiple cameras; the model is established by adopting a minimum cost flow network, the minimum cost flow network G comprises a node N, anedge E and a weight W, the weight W represents the value of the edge connecting the two nodes, and the nodes comprise a trajectory segment li node, specifically comprising a trajectory segment start node ienter and a trajectory segment end node iexit; the edge from the node ienter to the node iexit is an observation edge ei; the edge from the node iexit to the node jenter represents the connectionrelationship of the two track segments, and is a transfer edge eij; and the nodes further comprise a virtual source node S and a collection point node T. The start of the trajectory segment li and the end of the trajectory segment lj are represented by an input edge eSi and an output edge ejT respectively. According to the invention, group target trajectory analysis is realized.

Description

technical field [0001] The invention relates to a group target trajectory analysis model based on an overall graph averaging model. Background technique [0002] Object trajectory tracking has been a key problem in the field of computer vision for a long time. In recent years, many excellent apparent similarity algorithms have been proposed for single-camera visual tracking. However, in the intelligent monitoring system, it is difficult to capture the complete track of the apparent similarity of a specific target in a large-scale scene by using only one camera. Therefore, in practical applications, it is often hoped that the trajectory of a specific target with apparent similarity in a large-scale space can be obtained through multi-camera tracking. In the actual scene, there is often no overlapping image area between the apparent similarities of multiple adjacent cameras installed, which is indeed a great challenge for the apparent similarity of tracking the complete traj...

AI Technical Summary

Problems solved by technology

The problem of the traditional method is obvious. To solve the problem of ICT, the apparent similarity depends on the result of SCT. That is to say, the performance dependence on single-camera target tracking is too strong, which often leads to the overall tracking effect of the apparent similarity is very unstable.

[0004] The current SCT method is not robust enough, and its tracking results often contain many trajectory fragments and wrong trajectories, which is a fatal problem for ICT models

These trajectory fragments and wrong trajectory interference cause many problems in the ICT model, such as: 1) the problem of target mismatch

The track B in the second camera is incorrectly associated with the track A2 in the first camera, resulting in an error in the final track analysis result; 2) The problem of the loss of the target track

[0007] But the association of inputs (trajectory fragments) brings new problems: how to measure the similarity between two inputs

As a result, inputs from the same camera are often more easily correlated than inputs from different cameras, causing cross-camera trajectory tracking to fail

Images