Track prediction method and device based on time attention convolutional network

Abstract

The invention discloses a trajectory prediction method and device based on a time attention convolutional network. The method comprises the following steps: acquiring trajectory data of at least one target in a previous time period; inputting the trajectory data into a trajectory prediction model, and outputting a predicted trajectory of the target in the next time period by the trajectory prediction model; wherein the trajectory prediction model extracts time features of different scales of trajectory data through the time attention module and the first causal convolution module, and obtains a prediction trajectory of the target in the next time period based on the extracted time features of different scales. Integrating the influence of all moments in the previous time period of the target on the current moment through a time attention module, and automatically paying attention to historical time sequence characteristics with relatively large influence; processing the trajectory data in parallel through a first causal convolution module and generating corresponding time sequence feature data with the same length; the multi-scale time features of the trajectory data are quickly and accurately obtained, so that the obtained predicted trajectory of the target in the next time period is more accurate, and the obtaining speed is higher.

Description

technical field [0001] The present invention relates to the technical field of target moving trajectory prediction, in particular to a trajectory prediction method and device based on temporal attention convolution network. Background technique [0002] In the prior art, it is necessary to predict the trajectory of movable objects such as intelligent robots, automobiles, unmanned intelligent vehicles, and pedestrians. Usually, the trajectory prediction of the target in the next period is performed based on the trajectory sequence data of the target in the previous period. However, the trajectory sequence Processing has always been the difficulty of target trajectory prediction, that is, how to make full use of the currently observed sequence data and extract its features effectively. [0003] In recent years, the research on trajectory sequence processing has made great progress, and various effective models have been used in it. For example, the model commonly used in the p...

AI Technical Summary

Problems solved by technology

Because although the recurrent neural network intuitively conforms to the processing idea of ​​time series, the traditional recurrent neural network cannot process time series data in parallel.

The input at the current moment depends on the hidden state output at the previous moment, so it is lacking in model training and inference efficiency

Moreover, in the process of forward propagation, the cyclic recursive network constantly chooses to forget the characteristic knowledge of historical moments. Although the long short-term memory network claims to be able to process data of longer time series, the actual effect is not ideal.

In addition, when target trajectory prediction is applied to traffic scenarios, real-time prediction is usually required, and the recurrent neural network does not meet the requirements in terms of prediction speed

Images