Probabilistic Image-Driven Traffic Modeling via Remote Sensing

TL;DR

This work introduces a multi-modal, multi-task transformer-based segmentation architecture that can be used to create dense city-scale traffic models and evaluates the method extensively using the Dynamic Traffic Speeds dataset and significantly improves the state-of-the-art.

Abstract

This work addresses the task of modeling spatiotemporal traffic patterns directly from overhead imagery, which we refer to as image-driven traffic modeling. We extend this line of work and introduce a multi-modal, multi-task transformer-based segmentation architecture that can be used to create dense city-scale traffic models. Our approach includes a geo-temporal positional encoding module for integrating geo-temporal context and a probabilistic objective function for estimating traffic speeds that naturally models temporal variations. We evaluate our method extensively using the Dynamic Traffic Speeds (DTS) benchmark dataset and significantly improve the state-of-the-art. Finally, we introduce the DTS++ dataset to support mobility-related location adaptation experiments.

Figures (14)

• Figure 1: We propose a method for image-driven traffic modeling, which can be used to create dense city-scale traffic models. (left) Historical ground-truth traffic data is often sparse as not all roads are traversed at all times. For example in Brooklyn on Monday at 8am, many roads are missing empirical speed data. (right) Our method can create a dense model of traffic flow at the same time.
• Figure 2: An overview of our architecture for image-driven traffic modeling.
• Figure 3: An overview of our proposed geo-temporal positional encoding module.
• Figure 4: Example images from the Dynamic Traffic Speeds (DTS) dataset and corresponding labels. The right four labels depict available historical traffic speeds at different times, where green (red) corresponds to faster (slower) speeds.
• Figure 5: Qualitative results for traffic speed estimation (Monday, 8am). (top) Ground-truth traffic speeds are provided as aggregates for individual road segments. For visualization, we replicate the ground-truth speed across the entire road segment. (middle) Predictions from our approach, after applying region aggregation. (bottom) Our results without region aggregation capture nuances of traffic flow, such as slowing down around curves. Green (red) corresponds to fast (slow) traffic speeds.