Road detection method based on SIFT-COF feature optical flow

Abstract

The invention discloses a road detection method based on an SIFT-COF feature optical flow. Firstly, frame image information on a road is collected, and a road area and a non-road area are distinguished; the non-road area is determined to be ROI, then features in the ROI are extracted, and feature extraction is a hierarchical structure in which the SIFT feature and the Harris angular point feature are combined; moreover, the feature optical flow is formed through feature matching among frames, and a feature position is judged according to the calculation of the feature optical flow; finally a passable area and a non-passable area are determined. The hierarchical structure with the SIFT feature and the Harris angular point feature is built, complementary advantages of the SIFT feature and the Harris angular point feature are achieved through the method, and through the scale invariance character of the SIFT feature and the uniform distribution character of the Harris angular point feature, optical flow detection contrast tests of different types of unstructured roads show effectiveness of the method.

Description

technical field [0001] The invention relates to the field of autonomous navigation of intelligent vehicles, in particular to a SIFT-COF characteristic optical flow detection method for realizing the accuracy of road detection. Background technique [0002] According to unstructured roads, the vast majority of road areas have similar characteristics and have characteristic differences from non-road areas. Therefore, combining road region segmentation with road boundary detection methods can improve the accuracy and robustness of road detection. However, due to the different material composition of the road (such as mixed gravel, mud or sand) and the influence of environmental illumination, the local area of ​​the road will have characteristics different from other areas of the road itself, which is reflected in the non-road area during the road detection process. It is necessary to distinguish whether these non-road areas are roads or obstacles. Combining non-visual sensor ...

AI Technical Summary

Problems solved by technology

Combining non-visual sensor data with visual data can effectively overcome this problem, but there are the following difficulties: Unlike visual sensors that can theoretically see infinity, non-visual sensors may be due to detection angle problems (such as single-line lasers) The existence of obstacles cannot be detected, and it may only be able to detect short-distance obstacles due to the detection range

Using expensive 64-line lidar can detect medium-sized vehicle obstacles of about 100m, but its price is much higher than the cost of ordinary vehicles

Images