A method and device for predicting the position of a moving target under a fixed camera

Abstract

The present invention provides a method and device for predicting the position of a moving target under a fixed camera, wherein the method includes: acquiring data to be predicted, sending the data to be predicted into a neural network model for target position prediction, using the The neural network model predicts the data to be predicted. The data to be predicted includes video data and the time interval t to be predicted. The video data includes K frames of RGB 3-channel pictures; the K frames of RGB 3-channel pictures and the time interval t to be predicted are combined into 4 channel K×H×W size stereo; send the 4-channel K×H×W size stereo into the 3D deep convolutional residual network, extract features, and output feature maps of different scales; use feature pyramids for different scales The features are fused to obtain a 2D multi-channel feature map; using the 2D multi-channel feature map, the target position of the layer to be predicted is predicted, and the target position prediction result is obtained.

Description

technical field [0001] The invention relates to the field of computers, in particular to a method and device for predicting the position of a moving target under a fixed camera. Background technique [0002] The prediction of the position of the moving target is the key technology of the blind-guiding equipment. The blind-guiding equipment such as the blind-guide glasses is usually equipped with a monocular camera. How to locate the target and predict the position of the target from the video obtained by the monocular camera is very important for its own state estimation and path planning. important. [0003] The existing moving target position prediction technology is usually divided into two stages. Firstly, video target detection and target tracking algorithms are used to obtain the current and historical position of the target, and then predictive models such as recurrent neural network or tree search algorithms are used to predict the future trajectory of the target. T...

AI Technical Summary

Problems solved by technology

The disadvantage of these methods is that the rule-based method is difficult to enumerate all situations, while the cyclic neural network method is usually restricted by the capacity of the model, and end-to-end training cannot be performed, and the accumulated errors lead to unstable predictions and affect the prediction accuracy.

Images