Stereo odometry based on local intensity order pattern

New generation autonomous vehicles use different data fusion techniques to solve the Simultaneous Localization And Mapping (SLAM) problems in urban terrains. However, the majority of the implementations uses high-cost sensors like LIDAR to obtain a high accuracy map. In this paper, we present a method to solve this problem using sequences of stereo images. Our approach uses Local Intensity Order Pattern (LIOP) based feature descriptors to overcome the problem of monotonic intensity changes and affine transformations of detected features. Ego-motion of the vehicle is estimated using correspondence of features on the consecutive frames. Estimated motion parameters are then corrected using an extended Kalman filter. Depth of the detected features is calculated using stereo triangulation. Mahalanobis distance is used to avoid outliers in the detected features. The accuracy of the proposed method is evaluated using the measurements from a high accurate inertial navigation system. Method is tested using the odometry data set of the KITTI Vision Benchmark Suite.