Design an‎d Implementation of a Fast an‎d Accurate Stair Detection System for Mobile Robots Based on Depth Data Processing an‎d Deep Learning

10/9/2025 12:00:00 AM

With the advancement an‎d widespread application of mobile robots, their locomotion challenges, especially when encountering stairs, have gained increasing importance. Despite numerous studies in this field, the precise detection an‎d differentiation of stairs remains an open problem in robotics an‎d computer vision. This research presents a novel, efficient, an‎d fast algorithm for detecting an‎d differentiating various types of stairs, capable of performing well in a diverse range of environmental conditions (light levels, stair structures, etc.). The proposed method is designed based on processing depth data obtained from an RGB-D camera equipped with an IMU sensor. In this approach, depth data is first converted into an elevation map image using the IMU. Then, using a trained neural network, the staircase area is identified in the elevation map. In the next stage, the depth data within the detected area is processed an‎d clustered. After several stages of examining an‎d refining these clusters, the exact position of each step is determined. To visualize the output of this algorithm, a color mask is applied to the stairs in the RGB image for user display. Statistical eva‎luation results show that the proposed algorithm, in stair detection frames (which have the highest processing load), has an average cycle time of 0.252 seconds. Additionally, the accuracy of the neural network in detecting the precise location of the staircase has been eva‎luated with an average confidence index of 92.13 percent.

تشخيص پله , پردازش سريع , دوربين RGB-D , نقشه ارتفاع , يادگيري عميق

Staircase Detection , Fast Processing , RGB-D Camera , Elevation Map , Deep Learning

MohammadReza Nili

Dr. Mohammad Shahbazi