首页|A LIDAR-based Traffic Data Classification Framework for Indian Urban Traffic
A LIDAR-based Traffic Data Classification Framework for Indian Urban Traffic
扫码查看
点击上方二维码区域，可以放大扫码查看
原文链接
NETL
NSTL
Springer Nature
With recent advances in autonomous vehicles and traffic monitoring systems, the use of light detection and ranging (LIDAR) is becoming more popular. One of the essential components of these systems is a LIDAR point-cloud classifier. This work introduces a generalized classification framework based on traditional 3D point cloud processing algorithms, together with a classification model with interpretable inputs. The framework consists of three stages, wherein the first two stages involve the development of input features of the classification model through preprocessing and feature generation algorithms. In the final stage, the multiclass machine learning (ML) model predicts the vehicle class. The study also presents the refining of data from off-line techniques to improve the performance of the ML model. The framework is validated using real-world LIDAR data that represent heterogeneous laneless traffic. A comparison of a range of point-cloud ground segmentation and clustering algorithms is conducted on this data set, and it is shown that density-based spatial clustering of applications with noise (DBSCAN) and ground segmentation by m-estimator sample consensus (MSAC) give the best clustering output. Seven features representing the dimension, distribution, and density of the clusters were extracted using bounding-box fitting and line-fitting algorithms. After training with various classification models using these features, the adaptive boosting algorithm (AdaBoost) was determined to have the highest accuracy (0.922 F1 score) for five output classes. Furthermore, it is demonstrated that this accuracy can be enhanced by data-refining techniques such as skewness reduction and region-ofinterest boundary selection. The final model obtained has an accuracy of 98.4% (0.969 F1 score). The results show that the framework is well-suited for applications that employ multiclass classifiers for heterogeneous and laneless traffic.
LIDARTraffic data classificationMSACDBSCANAdaBoost
Prajwal Shettigar J、Arun K Tangirala、Lelitha Devi Vanajakshi
展开 >
Department of Data Science and Artificial Intelligence, Robert Bosch Centre for Data Science and Artificial Intelligence, Indian Institute of Technology Madras, Chennai, India
Department of Civil Engineering, Robert Bosch Centre for Data Science and Artificial Intelligence, Indian Institute of Technology Madras, Chennai, India
NETL
NSTL
Springer Nature
