Rapid track prediction method for large-scale moving object, medium and equipment

Abstract

The embodiment of the invention provides a rapid track prediction method for a large-scale moving object, a medium and equipment. According to the invention, the pressure of large data volume is relieved through track compression, soft rewards and punishment measures are adopted, that is to say, after matching succeeds, the mean value of the Gaussian probabilities of two corresponding adjacent historical rule points is used as a similarity reward to prevent reward and punishment measures from being too sharp, so that track prediction accuracy is improved. After the compressed historical regular track with the highest similarity with the compressed current track is obtained, the real-time performance is improved through an interpolation completion method under the condition that the track prediction accuracy is not reduced.

Description

technical field [0001] The present invention relates to the field of information technology, in particular trajectory prediction technology, and more specifically, a method, medium and equipment for rapid trajectory prediction of large-scale moving objects. Background technique [0002] Real-time trajectory prediction is to predict the moving path of the target in the future. For individuals, if their movement path in the future can be effectively predicted, personalized service pushes can be made, for example, push reminder services, advertisements, etc. For groups, accurately predicting the future movement path of each target can facilitate intelligent traffic management, for example, providing intelligent traffic services based on trajectory prediction. [0003] The usual implementation process of trajectory prediction is to calculate the similarity between the current trajectory of the target and all the historical regular trajectories of the target, and then predict th...

AI Technical Summary

Problems solved by technology

[0009] There are two problems in the above trajectory similarity measurement algorithm: the first problem is that in the reward and punishment measures for calculating the similarity, either directly use the absolute distance between two points, or directly add 0 or 1, this kind of reward and punishment measures are too large sharp, which widens the gap between similarity and dissimilarity, and reduces the accuracy of trajectory prediction; the second problem is that the real-time performance of trajectory prediction is poor, because in the case of large-scale moving objects, the group target for trajectory prediction The amount of data is large, the number of rules of a single target is large, and the time complexity of the similarity measurement algorithm itself is O(n 2 )(where n represents the total number of current trajectory points, O(n 2 ) means that the algorithm will take n time to complete the operation 2 ), which leads to the situation of large-scale moving objects, the amount of trajectory data (ie n) is large, and there must be a similarity measure in the step of trajectory prediction, and the time complexity of the similarity measure algorithm is O(n 2 ), the duration of trajectory prediction will become larger, reducing the real-time performance of trajectory prediction

Images