Research On Lane Line Detection Mathod Based On LIDAR

As the next blue ocean which will change the way people travel,unmanned technology is becoming a hot research at home and abroad.A powerful sensing system is the basis for safe driving of unmanned vehicles.When an unmanned vehicle is driving on a structured road with lane markings,the lane markings are usually the highest priority reference.Therefore,lane recognition is an important factor in determining autopilot performance.Intelligent vehicleassisted safety systems and autonomous navigation systems can only play an important role if the vehicle’s lane and lane position are accurately detected.Currently,the two main methods of lane line detection are based on vision detection and lidar detection.Among them,the development of lane line detection technology based on machine vision is relatively mature.However,under poor lighting conditions in the external environment,such as strong light,night and shadow,it is difficult to obtain satisfactory results.Existing methods can partially improve the recognition rate,but the problem can not be solved fundamentally,because the quality of sensor data is always the most important factor.With the continuous development of sensor technologies such as radar,the lidar based lane detection method has become a research hotspot.This method is less affected by the environment,which makes up for the lack of camera.Moreover,the detection accuracy of the lidar is high,and the depth information of the environment around the road can be obtained.An approach of lane detection based on LIDAR is proposed in this thesis,Firstly,the position of the road curb is detected,and the effective data of the road surface where the lane line is located is divided by the position of the road edge.Then the position of the lane line is detected from the road surface data according to the lane line characteristic.The specific work mainly includes the following two aspects: Firstly,the road curb detection stage: First,the RANSAC fitting plane method is used to roughly extract the road area,filtering out most of the non-ground data and improving the processing speed of the subsequent steps.Then,an extraction algorithm based on the unrelated graph neighborhood relationship is proposed in this thesis,which is characterized by multi-feature,wide-threshold curb feature.By setting a variety of roadside geometric features to set a wider threshold to improve the accuracy of the edge detection.Secondly,lane detection phase: a local adaptive threshold algorithm is proposed to extract lane features based on the distinct difference of the reflective intensity between lane and road surface,which avoids the problem that the global threshold can not balance the situation in the point cloud image that resulting in poor lane feature extraction.A roadside and laneline clustering algorithm combining density and continuous characteristics in the direction of the road is proposed.The least squares method is used to realize the fitting of the roadside and lane lines.The research in this thesis is tested by collecting road environmental data.The results show that the proposed method can accurately detect the lane lines under different road environments and different illumination conditions,and the detection accuracy is improved.