Abstract:
The primary roads of Dhaka are essential infrastructural components, similar to those in cities around the world. These roads facilitate major traffic movements within urban centers and between key municipal traffic generators. The operating speed of these roads has a significant impact on their performance. Speed is a crucial factor in urban primary roads, as it affects the density and volume of traffic streams. Recent studies have focused on developing speed prediction models for homogeneous traffic conditions on two- or multi-lane rural highways. However, there is a lack of research on urban primary roads under heterogeneous traffic conditions.
In cities like Dhaka, there is a notable difference in behavior between heterogeneous and undisciplined traffic streams compared to conventional traffic flow. In Dhaka, traffic streams are mixed, with various vehicles occupying the same lane and moving freely wherever space permits. Thus, the primary road system in Dhaka warrants an in-depth study and the development of a speed prediction model to support efficient roadway management. This study aims to create simultaneous speed prediction models based on fundamental speed-density relationships in heterogeneous traffic conditions.
The basic Greenshield’s equation, which relates speed to density, has been modified for mixed traffic compositions and the proportion of different vehicle types within the traffic stream. However, traditional fundamental speed-density models with the least squares fit method often struggle due to their few limitations. We applied the Maximum Likelihood Estimation (MLE) technique and Bayesian Inference to address these limitations. We selected a 400-meter section along Mirpur Road, Matikata Road, Airport Road, Toyenbee Circular Road, and Darus Salam Road, all of which exhibit uniform traffic operating conditions without access points. A longitudinal trap of 60 meters was set up in the middle of this section to measure flow and speed. Data was collected using a DJI AIR 3 drone equipped with a dual-camera system and 4K/60fps HDR video on a typical weekday from 8:00 a.m. to 6:00 p.m. Vehicles in the traffic stream were categorized into six groups: Standard Cars (SC), Utility Vehicles (UV), Heavy Vehicles (HV), Three-Wheelers (3W), Two-Wheelers (2W), and Non-Motorized Vehicles (NMV).
A deep learning-based heterogeneous traffic measurement technique was utilized to analyze and detect vehicles. Vehicle trajectories were tracked using DeepSORT, while vehicle identification and classification were performed using YOLOv8. This study employed DEEGITS (Deep Learning Based Heterogeneous Traffic State Measurement), a comprehensive framework that utilizes YOLOv8 and DeepSORT for accurately detecting, classifying, and tracking vehicles. Several scripts were written in MATLAB to analyze the processed video data and extract relevant attributes, including vehicle type, speed, and flow rate, for each road section. Based on a comprehensive analysis of the traffic data collected from the selected urban primary road network segments, this study successfully developed a simultaneous speed prediction model to estimate vehicle speeds under heterogeneous traffic conditions.
Using the developed model, the analysis thoroughly investigated traffic flow rates for different Dhaka City traffic stream compositions. Initially, with a vehicle mix of Standard Cars (SC), Utility Vehicles (UV), Heavy Vehicles (HV), three-wheelers (3W), two-wheelers (2W), and Non-Motorized Vehicles (NMV) at 30%, 10%, 10%, 10%, 10%, and 30%, the total vehicle volume showed better performance, but it did not maximize road capacity. A mix of 20%, 10%, 30%, 10%, 20%, and 10% yielded satisfactory results in an analysis at 4500 vehicles per hour. However, a composition of 25%, 25%, 15%, 15%, 15%, and 5% reached critical density at 6500 vehicles per hour. Finally, the optimal mix of 20%, 10%, 30%, 15%, 15%, and 10% demonstrated the best speed and roadway capacity optimization. The accuracy of the equations is checked by comparing the observed values of speed with the calculated speeds.
The models developed in this research will enable transportation engineers to plan more effectively and improve road operation and safety management. These models could be integrated within Intelligent Transportation Systems (ITS) and utilized with Variable Message Signs (VMS) across Dhaka's primary road network. To achieve this, ITS would need to gather extensive data from UAVs monitoring the roadway and continually process the information using the established prediction models.