Drivable Area Detection By Multi-sensor Data Fusion

As an important part of the environment perception,the detection of drivable areas is the key to the safty of unmanned vehicles.It can detect road regions and boundaries while providing information for other modules,such as the planning module and the control module.This dissertation focuses on the detection of drivable areas including the structured and semi-structured road based on lidar and the fusion of multi-sensors.1.This dissertation builds a multi-sensor platform consisting of two cameras,a lidar,an inertial measurement unit(IMU)and global navigation satellite system(GNSS).Meanwhile We also perform calibration to obtain the transformation matrices for data fusion to ensure the unmanned vehicle's safety and reliability.2.This dissertation proposes a lidar-imagery based road detection method.We first construct a three-dimensional histogram of the point clouds according to the characteristics of the 64-channel lidar data,and extract the flat region using random sample consensus(RANSAC).Then,a quadric surface model is applied to fit the road region,followed by the analysis of each scan line separately to achieve a finer result.3.This dissertation proposes a road detection method based on the fusion of visible light images and point cloud data.We first project the sparse point clouds onto the image plane based on the calibration results,and then upsample them to generate the dense lidar image by checking whether each pixel is an edge pixel.Further,we design an end-to-end detection network to perform pixellevel segmentation of the drivable region by fusing the information between the two modalities to realize a reliable detection of the all-weather road areas.