Technology Overview
Landsonar aggregates and enhances historical and real-time traffic data from a vast number of private and public sources. We aggregate these disparate sources and types of heterogeneus data, normalize and convert it to Landsonar’s vector data format, apply proprietary predictive algorithms to determine road speed and time estimates, format these predictions as map data using standard web services and transportation management system protocols, and distribute our traffic information to end-users via a software service or embedded in third-party products.
Our predictive traffic information system is built on an aggregation of historical traffic data as well as on data regarding recurring and upcoming events that may affect traffic congestion and road speed. Data is collected from a variety of sources, including “probe vehicles” (typically, GPS receivers of various kinds mounted in many kinds of vehicles), fixed roadbed or roadside sensors (primarily public sources), and other non-road data. This traffic data is then enhanced by incorporating ancillary sources such as real-time traffic incident logs, weather reports and forecasts, road construction data, public events, etc.
The Computational Process
The first phase of our computational process is to extrapolate road speeds from latitude/longitude and other information collected by GPS-equipped vehicles, phones, and other sources; and to map these speeds, along with speed data from other sources, into digital map segments. A map segment is the most granular building block of a digital map, each segment typically representing between 250 and 1,400 feet of travel distance. Traffic patterns must be modeled at this highly granular level to be truly effective. This process takes into account an understanding of the idiosyncracies of each data source and how they compare to each other, and incorporates significant error correction and detection elements. We have learned these idiosyncracies the hard way — through hands-on experience in manipulating, modifying, and marrying more source data than anyone else ever before.
At this point in the process, the road speed data is correlated to tertiary data and evaluated for patterns. Forecast speeds based on these patterns are then calculated using one of several different predictive algorithms. Multiple algorithms are used simultaneously and later crosschecked against actual recorded speeds for improvement via a process of feedback. Depending on the needs of the end-user, these forecasts can be as simple as averages or as complex as a probabilistic distribution. Depending on the nature of their deployment, end-users can also select near-term forecasts (12-48 hours) for server-based deployments, or long-range forecasts (3-12 months) for embedding into end-user products.
The granular nature of LandSonar’s data output is designed to facilitate distribution and integration with the digital mapping systems found at the core of all navigation and routing systems. All predictive traffic data is output as a map segment road speed by time of day, with a speed and a probability prediction for each segment. There are two dominant map vendors in the U.S. (NAVTEQ & Tele Atlas) and their maps have up to 40 million map segments. LandSonar is formatting its data according to the proprietary map segment ID tags (also called “link IDs”) of each map vendor and has built predictive road speed models for the entire road network in both map formats.
Once the data is formatted according to map vendor segments, it is then distributed in standard web services format to our customers and partners. Virtually all navigation and routing systems available today and in the foreseable future use the same digital map format and therefore are candidates for LandSonar’s traffic information. Much like the map data vendors, LandSonar provides different aspects of the same data to different markets, with pricing, value proposition, and channel distribution as appropriate for each target market.
