Vehicle Ego-Localization Based on the Fusion of Optical Flow and Feature Points Matching

To meet the requirement of vehicle real-time and precise ego-localization on the flat road of city, a vehicle ego-localization method based on the fusion of optical flow and feature points matching is proposed. A novel FAST algorithm with self-adaptive threshold is applied to detect feature points. Based on the assumption of flat plane, the improved Lucas-Kanade algorithm is carried out to track feature points, and then the custom LARSAE is used to amend vehicle offsets. Meanwhile, Hu moments are used as the feature descriptor to complete image matching, realizing vehicle motion estimation. These two methods are fused by the discrete kalman filter to update and optimize vehicle position. Experimental results show that the fusion algorithm overcomes the shortcomings of poor positioning accuracy of optical flow and the low processing speed of feature matching, and is able to provide more accurate real-time positioning output, having a certain robustness for circumstances such as illumination change and low pavement texture.