Method of trajectory clustering based on directional trimmed mean distance

Abstract

The invention discloses a method of trajectory clustering based on directional trimmed mean distance (DTMD). The method comprises the following steps of: (1) trajectory extraction: extracting the trajectory from an original dynamic video sequence by using a motion tracking algorithm; (2) trajectory pretreatment: pretreating the extracted trajectory to reduce influences of situations of incomplete trajectory caused by missed tracking, false tracing, sheltering and the like during target tracking or noise point pollution and the like on consequent treatments; (3) similarity degree computation: computing similarity degrees among trajectories by utilizing a DTMD similarity degree formula and constructing a similarity degree matrix; (4) spectrogram clustering: converting the trajectories and similarity relationships thereof into a weighted graph, wherein an apex of the graph stands for the trajectory, edges stand for the similarity degree among corresponding trajectories, computing a characteristic root and a characteristic vector of the similarity degree matrix by utilizing a Laplace equation, and segmenting the graph by utilizing a Fielder value; and (5) clustering result obtaining: converting the segmented result of (4) into trajectory classification, marking the original trajectory and outputting the trajectory clustering result.

Description

technical field [0001] The present invention relates to a trajectory clustering method based on directional trimmed mean distance (DTMD, Directional Trimmed MeanDistance) (hereinafter referred to as the trajectory clustering method based on DTMD), which relates to a trajectory clustering method of moving objects, especially a Trajectory clustering method based on directed truncated mean Hausdorff distance. Background technique [0002] The task of video surveillance is to monitor the motion of objects in a scene. With the development of video surveillance technology, video surveillance has developed from simple low-level detection and tracking to classification of moving objects, and gradually extended to advanced behavior analysis using scene knowledge structure. That is to say, the scene knowledge is used to describe the behavior through the interactive relationship between space and time between objects, such as vehicles leaving the road, pedestrians waiting for the bus,...

AI Technical Summary

Problems solved by technology

[0005] The trajectory difference measurement based on Euclidean distance is very sensitive to the timing change of the flow vector and is easily affected by tracking noise

In the trajectory clustering method based on the longest common subsequence, it is difficult to determine the spatio-temporal error threshold parameters, there is no effective selection basis for reference, and these thresholds are fixed constants, lacking in adaptability

Although the traditional Hausdorff distance is easy to implement in calculation, in the process of solving, it is necessary to compare the distance between each point on each trajectory and each point on all other trajectories to finally obtain two trajectories. The distance between the traditional Hausdorff distance makes the calculation of the traditional Hausdorff distance larger; it is also very sensitive to noise and occlusion, and the existence of out-of-grid points often causes misjudgment; and because the traditional Hausdorff distance is a method for judging the distance between any two sets method, so no directionality is involved

In the clustering of trajectories, the trajectories are oriented. For example, the shape of the two trajectory line segment motion trajectories is basically the same and the distance is very close, but the direction is opposite. When using the traditional Hausdorff distance for clustering, it is often clustered into the same class, then serious errors may occur when the clustering results are used to classify later, such as vehicles going against the road in the lane will be classified into the same class

Images