Staircase Identification And Structural Parameter Estimation Based On Multi-directional One-dimensional Depth Feature

Environment perception is an important task in computer vision.It has important research significance in mobile robot autonomous navigation and guide blind application.Environmental perception mainly includes three aspects:roads,static objects and dynamic objects.In a complex road environment,stairs as a special scene is one of the main difficulties in environment-aware applications.Stair scene recognition has experienced the development from two-dimensional planar image to three-dimensional depth.The traditional two-dimensional planar image-based staircase recognition technology uses stair edge lines and texture features as the basis for identification.Although it is simple and rapid,it depends on the edge line of the stairs.The recognition accuracy is susceptible to similar types of features,such as zebra crossings.In the three-dimensional image,the depth information of the environment is obtained by the laser sensor or the binocular camera,and the structural features of the stair space are restored,which is suitable for three-dimensional reconstruction and recognition of the environment.However,the direction of the line laser is single,the 3D laser scanning is time-consuming and costly,and the binocular camera simulates the human eye to process the environmental information effectively and reliably,which has become a research hotspot for the scene detection of stairs.However,there are still many challenges in the current scene detection based on binocular stereo vision:1)Stair integrity issues.Stair material,edge structure,ambient light,noise,obstacles,etc.will cause the edge of the stairs in the image to be incomplete,affecting the recognition results;2)the problem of stairway orientation.During the movement,the posture change and rotation will lead to changes in the observation angle,which will cause the image of the stairs in the field to be inclined and reduce the recognition rate.In response to the above challenges,this paper proposes a stair scene recognition method based on multi-directional one-dimensional depth features.The method extracts the environmental one-dimensional depth information based on the binocular stereo vision system,and uses the iterative least squares method to fit the feature parameters,and uses the one-dimensional depth feature parameters as the classification basis to construct a multi-classification model,which is output by probability weighting method.Final classification results.The one-dimensional depth feature characterizes the three-dimensional structure of the stairs,and has nothing to do with the edge characteristics of the stairs,which can effectively solve the problem of stair integrity;multi-directional one-dimensional depth feature extraction can solve the problem of stairway directionality,and is applied to the identification of oblique stairs,which has wider application range.In order to achieve accurate scene detection,this paper estimates the structural parameters of the stairs on the basis of stair recognition,including the height and depth of the stairs,and builds the stair structure model through quantitative parameters.The point cloud library(PCL)is used to reconstruct the 3D environment,and the multi-plane extraction method of the staircase is proposed.The structural parameters of the up and down stairs are estimated based on the plane extraction and geometric model.This paper experiments and estimates various building stairs in the actual environment.The experimental results show that:1)in the identification experiment of non-complete stairs,the recognition accuracy is 94.54%;in the oblique staircase recognition experiment,the recognition accuracy is 90.94%.In the up-and-down staircase classification experiment,the F1 values of the upper and lower stairs are 95.86%and 95.18%,respectively.2)In the experiment of stair parameter estimation validity,the parameter estimation method is effective for the first three stages of the stairs;in the estimation accuracy experiment,the overall height estimation error is 5.30%,the depth estimation error is 4.81%,and the plane estimation error is 2.03%.The recognition method of stair recognition based on multi-directional one-dimensional depth feature in this paper is better than traditional feature recognition method in non-complete staircase and oblique staircase.Stair structure parameter estimation method has high precision and small error,which is used for stair scene recognition.And modeling provides new ideas and technical methods.