A method and system for estimating vehicle speed based on neural network

Abstract

The invention discloses a neural network-based vehicle speed estimation method and system. The vehicle speed estimation method includes obtaining training samples and training output, and expanding the training input in the training sample into an 8*8 symmetrical vehicle real-time data matrix, wherein Both the training output and the training input are in the form of vectors; then the convolutional neural network is trained according to the symmetric vehicle real-time data matrix and the training output; finally, the real-time data of the current vehicle is obtained, and the real-time data of the current vehicle is input into the post-training In the convolutional neural network model, the lateral velocity and longitudinal velocity of the current vehicle to the ground are estimated. The present invention utilizes a convolutional neural network, takes the symmetrical vehicle real-time data matrix as input, and outputs the real-time longitudinal and lateral speeds of the vehicle to the ground through convolution calculation convergence. And it effectively solves the cumulative error problem in the later integration process.

Description

technical field [0001] The invention relates to the technical field of vehicle speed estimation, in particular to a neural network-based vehicle speed estimation method and system. Background technique [0002] Based on the collected value of the acceleration sensor, the unscented Kalman method is used for filtering prediction, and the acceleration value is integrated to obtain the corresponding lateral velocity and longitudinal velocity. The disadvantage is that the unscented Kalman method has a large amount of calculation and cannot effectively solve the later integration process. The cumulative error problem in . [0003] Based on the collected four-wheel wheel speed signals, combined with the four-wheel model of the vehicle, the calculation of longitudinal speed and lateral speed has good real-time performance, but its disadvantage is that the model is established under the condition of no slippage of the vehicle, so it is not suitable for Extreme conditions of vehicle ...

AI Technical Summary

Problems solved by technology

[0002] Based on the collected value of the acceleration sensor, the unscented Kalman method is used for filtering prediction, and the acceleration value is integrated to obtain the corresponding lateral velocity and longitudinal velocity. The disadvantage is that the unscented Kalman method has a large amount of calculation and cannot effectively solve the later integration process. The cumulative error problem in

[0003] Based on the collected four-wheel wheel speed signals, combined with the four-wheel model of the vehicle, the calculation of longitudinal speed and lateral speed has good real-time performance, but its disadvantage is that the model is established under the condition of no slippage of the vehicle, so it is not suitable for Extreme condition of vehicle skidding

Images