Pedestrian detection and recognition using lidar for autonomous driving

In order to ensure the safe driving, autonomous vehicles under dynamic environments must accurately identify important targets that may change or move, especially pedestrians. Because the cameras are vulnerable to weather, light and other environmental factors, the real-time and accuracy of target detection and recognition may be poor. In this paper, we propose a method for pedestrian target recognition based on Lidar. It consists of three parts: ground point cloud removal, target object segmentation and pedestrian target recognition. Firstly, due to the huge amount of ground point cloud data, this paper uses a range image based on angle threshold algorithm to rapidly remove ground point cloud. With the mapping relationship between the disordered 3D point cloud and the ordered range image, the ground point cloud is marked and removed by the angle threshold on the range image rather than on the 3D point cloud. The ground point cloud removal is convenient for subsequent target object segmentation and recognition. Secondly, because the traditional clustering segmentation algorithm can not meet the real-time requirement of point cloud segmentation, we use improved Eps parameter DBSCAN algorithm combined with range image to segment point cloud target in this paper. For the problem of non-uniformity of 3D point cloud data density, we improve the setting of Eps parameter. For the undersegmentation between adjacent targets, we consider both Euclidean space distance and the angular distance which can distinguish the adjacent targets well in the space. Finally, considering the sparseness and disorder of the 3D point cloud and the lack of local feature extraction of the traditional PointNet network, a multi-scale feature fusion PointNet network is proposed, which combines the multi-scale local features and global feature of point cloud. The network also uses the spatial information and reflection intensity of Lidar point cloud to complete the pedestrian recognition. The experimental result shows that our proposed method can rapidly remove the ground point cloud and works well in segmentation between adjacent targets. The multi-scale feature fusion PointNet network performances well on 3D point cloud, its Area Under the Curve of ROC value reaches 0.92.

[1]  M. Himmelsbach,et al.  Real-time object classification in 3D point clouds using point feature histograms , 2009, 2009 IEEE/RSJ International Conference on Intelligent Robots and Systems.

[2]  Kyungeun Cho,et al.  A Fast Ground Segmentation Method for 3D Point Cloud , 2017, J. Inf. Process. Syst..

[3]  Dima Damen,et al.  Recognizing linked events: Searching the space of feasible explanations , 2009, 2009 IEEE Conference on Computer Vision and Pattern Recognition.

[4]  Csaba Benedek,et al.  Instant Object Detection in Lidar Point Clouds , 2017, IEEE Geoscience and Remote Sensing Letters.

[5]  Tian Xia,et al.  Vehicle Detection from 3D Lidar Using Fully Convolutional Network , 2016, Robotics: Science and Systems.

[6]  Ji Wan,et al.  Multi-view 3D Object Detection Network for Autonomous Driving , 2016, 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR).

[7]  Sebastian Scherer,et al.  VoxNet: A 3D Convolutional Neural Network for real-time object recognition , 2015, 2015 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS).

[8]  Diego Alonso,et al.  A Machine Learning Approach to Pedestrian Detection for Autonomous Vehicles Using High-Definition 3D Range Data , 2016, Sensors.

[9]  Jimmy Ba,et al.  Adam: A Method for Stochastic Optimization , 2014, ICLR.

[10]  Jun Miura,et al.  Pedestrian Recognition Using High-definition LIDAR , 2011 .

[11]  Li Wanlin,et al.  Target Segmentation Method for Three-Dimensional LiDAR Point Cloud Based on Depth Image , 2019 .

[12]  Leonidas J. Guibas,et al.  PointNet: Deep Learning on Point Sets for 3D Classification and Segmentation , 2016, 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR).

[13]  Yin Zhou,et al.  VoxelNet: End-to-End Learning for Point Cloud Based 3D Object Detection , 2017, 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition.

[14]  Leonidas J. Guibas,et al.  Frustum PointNets for 3D Object Detection from RGB-D Data , 2017, 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition.

[15]  Christoph Mertz,et al.  Pedestrian Detection and Tracking Using Three-dimensional LADAR Data , 2010, Int. J. Robotics Res..

[16]  Jian Cheng,et al.  Robust vehicle detection using 3D Lidar under complex urban environment , 2014, 2014 IEEE International Conference on Robotics and Automation (ICRA).

[17]  Wen-Huang Cheng,et al.  Multi-cue pedestrian detection from 3D point cloud data , 2017, 2017 IEEE International Conference on Multimedia and Expo (ICME).

[18]  Andreas Geiger,et al.  Are we ready for autonomous driving? The KITTI vision benchmark suite , 2012, 2012 IEEE Conference on Computer Vision and Pattern Recognition.