A LiDAR-based methodology for monitoring and collecting microscopic bicycle flow parameters on bicycle facilities

Research on microscopic bicycle flow parameters (speed, headway, spacing, and density) is limited given the lack of methods to collect data in large quantities automatically. This paper introduces a novel methodology to compute bicycle flow parameters based on a LiDAR system composed of two single-beam sensors. Instantaneous mid-block raw speed for each cyclist in the traffic stream is measured using LiDAR sensor signals at seven bidirectional and three unidirectional cycling facilities. A Multilayer Perception Neural Network is proposed to improve the accuracy of speed measures. The LiDAR system computes the headway and spacing between consecutive cyclists using time-stamped detections and speed values. Estimation of density is obtained using spacing. For model calibration and testing, 101 hours of video data collected at ten mid-block sites are used. The performance of the cyclist speed estimation is evaluated by comparing it to ground truth video. When the dataset is randomly split into training and test sets, the RMSE and MAPE of the speed estimation method on the test set are 0.61 m/s and 7.1%, respectively. In another scenario, when the model is trained with nine of the ten sites and tested on data from the remaining site, the RMSE and MAPE are 0.69 m/s and 8.2%, respectively. Lastly, the relationships governing hourly flow rate, average speed, and estimated density are studied. The data were collected during the peak cycling season at high-flow sites in Montreal, Canada; However, none of the facilities reached or neared capacity.