Scene recognition method for mobile robot based on esn neural network

Abstract

A mobile robot scene recognition method based on ESN neural network: extract ORB feature points from the image sequence that constitutes the scene, and then extract PIRF feature points from the extracted ORB feature points; point, use the bag-of-words model to establish a bag-of-words codebook, and use the bag-of-words codebook to calculate the encoding vector for each image in the image sequence; build an ESN neural network, use the encoding vector of each image as the input of the ESN neural network, and train online The ESN neural network identifies and extracts the scene information in the image sequence according to the output result. The present invention uses the randomness brought by dynamic objects in the scene to scene recognition, and has strong adaptability to high dynamic environments. The invention greatly reduces the search scale and difficulty, and improves the efficiency of search and identification. The invention has good applicability, can conveniently and effectively complete the identification and classification of scene sequences, ensures the accuracy of identification and improves the identification speed.

Description

technical field [0001] The invention relates to a scene sequence recognition method of a mobile robot. In particular, it relates to a mobile robot scene recognition method based on ESN neural network. Background technique [0002] Vision-based scene recognition refers to the process in which the robot uses the images collected by the visual sensor to allow the robot to judge and recognize whether the current scene has been passed through. In many vision-based scene recognition algorithms, the recognized scene is a single image, that is, matching between images is performed to complete the recognition. In order to solve this problem, people often start from these two aspects: the first design has discriminative image features, such as SURF, SIFT, MLSD and other descriptors; the second design recognition matching strategy, such as nearest neighbor, probability model, HMM Wait. Although these methods have achieved scene recognition and matching to a certain extent, due to th...

AI Technical Summary

Problems solved by technology

Although these methods have achieved scene recognition and matching to a certain extent, due to the variability of the environment, the influence of robot movement and other factors, a large number of misidentifications have occurred.

For example, several images collected continuously by the robot are very similar due to the speed of the robot's movement, which often has a great impact on the recognition

Images