Single-Layer Laser Scanner-Based Approach for a Transportation Participants Recognition Task
D-W. Gong, X. Dai, Y. Chen and S-F. Wang
The knowledge of quantities and types of traffic participants at specific nodes dedicates crucial information to an intelligent transportation system (ITS) of a smart city. A single-layer laser scanner was employed on a static platform on the roadside to recognize pedestrians, bicycles, motorcycles, trucks, sedans and buses. The selected features were extracted and investigated for optimized classification rate. The classifier support vector machine (SVM) is configured to classify the types of traffic participants using multiclass classification approach. To improve the recognition correction rate, the K-folder cross-validation and grid search are applied in the training stage of the machine learning. The hierarchical division of the output space is proposed where five classifiers were applied to deal with the specific difficult classes. The overall accuracy of the final experiment is 85.5% where the multiclass accuracies are 91.9, 99.6, 97.8, 96.6 and 80.4%, respectively, which shows that the laser scanner used is a feasible and practical solution of this application.
Keywords: Laser scanner, traffic participants, multiclass classification, support vector machine (SVM), intelligent transportation system (ITS)