Error modelling of depth estimation based on simplified stereo vision for mobile robots
Bo Jin, Lijun Zhao, Shiqiang Zhu
The State Key Lab of Fluid Power Transmission and Control, Zhejiang University, Hangzhou, 310027, China
Depth estimation is the precondition in obstacle avoidance for mobile robots. To improve the obstacle detecting effectiveness and quickness in poor-textured backgrounds, we used the centroid abscissa difference of corresponding obstacle region in image pairs as parallax to estimate obstacle depth. The error of parallax and depth were studied analytically and numerically. Wood blocks of different shapes and sizes were used for demonstrating the relationship between estimated depth and actual depth. A quadratic function model was obtained after experiments. Although the depth estimation error was relatively higher compared to conventional grayscale correlation-based method, the proposed method was expected to satisfy the accuracy requirement of depth estimation for common mobile robots.