Road Traffic Analysis Using 2D LIDAR

Abstract

Traffic congestion that leads to pollution and loss of valuable time and money of citizens is becoming a major concern. Understanding the congestion and developing mitigating measures is the need of the hour. Identifying key traffic variables for congestion quantification is pivotal. Volume, speed, and density are commonly utilized metrics in this regard. The current study uses a relatively new sensing technology, the Light Detection and Ranging (LIDAR) for analyzing traffic flow and congestion. A 2-D LIDAR system is specifically deployed at a selected location for this purpose. Density-Based Spatial Clustering of Applications with Noise (DBSCAN) algorithm is used for vehicle detection from the LIDAR data. The total vehicle count is estimated with an accuracy of 99%, and the estimation of classified vehicle count showed a mean absolute percentage error of 1.21%. The performance is evaluated with the help of field-collected video data. Road occupied area is also determined based on which congestion was estimated. Further, a forecasting model is developed and implemented using a stacked LSTM (Long Short- Term Memory) neural network to predict the next instants of occupied area, which gave a mean square error of 0.02 on the test data.