Solving Wide-Area Object Detection Challenges with Orbital Insight GO
We recently launched our Orbital Insight GO Platform, supporting DIY geospatial analytics at scale. A core feature of the platform is the ability for users to monitor activity-based intelligence. The theory of activity-based intelligence centers around finding human-made objects as proxies for nearby human activity and observing emerging trends including baseline and anomalous activity levels.
At Orbital Insight, we use the GO Platform to monitor and analyze these patterns by developing a variety of computer vision object detection algorithms that find human-made objects of interest. Searching across wide areas for these objects — even full country scale — GO users spend less time on data creation (e.g. a list of all planes in the area) and more time on data analysis (e.g. why are there more aircraft in this area than usual?). In this post, we’ve shared some example use cases of challenges the GO Platform can conquer for informed client decision making.
Difficulties of Obtaining Accurate Wide Area Data at Scale
Oftentimes, generating an accurate and recent dataset that spans an entire city, region — or even a whole country — can be a real challenge. For example, household surveys are difficult to scale to larger administrative levels in addition to being both time and resource intensive. Object detection at scale poses a challenge even for analysts with access to satellite imagery.
Analysts can spend days or weeks reviewing pixels from satellite imagery and creating data to answer a single question depending on the volume of imagery and number of objects present. Counting cars in urban areas can take a trained analyst around 156 seconds per kilometer squared, and the same task takes GO around .07 seconds per kilometer squared. These calculations are based on observing human analysts counting efforts and GO project status for an urban area car counting project. The efficiency estimates change depending on the object, imagery volume, and overall area object density (cities versus rural areas versus open ocean).
GO serves the user by generating data that is otherwise difficult to obtain, and does so much faster than the human analyst. We’ve used wide area object detection data generated by Orbital Insight’s GO Platform to generate insights that help clients answer complex questions:
- What are the impacts of ceasefire negotiations on economic activity in conflict environments?
- Have increased city-wide fuel taxes and public transit investments actually decreased the number of vehicles on the roads?
Orbital Insight GO detects objects by applying computer vision algorithms that scan entire catalogues of commercial satellite imagery and other geospatial data sources to find and count those objects. GO applies a user-centric approach, providing insights based on user criteria. Users can choose:
- Objects: “I want to count all ships across the city ports.”
- Dates: “I want to count all vehicles from 2014 to 2018.” Or “I want to count all vehicles from now on.”
- Areas: “I want to count all planes across South America.”
The GO Platform counts the objects based on a user’s input parameters, then analyzes and visualizes trends in activity or patterns of life. Users also set alerts to notify them when anomalous or unusual behavior occurs — such as more cars than usually observed on a Wednesday in the area — and can even set GO to kick off a new project to investigate the change in activity.
Economic Indicators in Yemen’s Conflict Environments
We used the wide area monitoring feature in Orbital Insight GO to count millions of passenger vehicles across major cities in Yemen — in just a few hours. This project supports clients seeking to understand the impact that both ceasefire agreements and violations had on economic activity.
Drops in the car detections observed in Yemen correspond to the dates of major attacks or fighting that occurred in those cities.
However, an increase in port traffic was observed at the same dates when rebels agreed to withdraw their control of the ports essential for the inflow of the country’s economic and aid resources.
This dataset was complemented by further analysis of ship detections at all of Yemen’s major ports where rebel forces recently gave up control in May 2019. Quantities of cars on roads or ships at ports are important economic indicators, especially when values are generated in near real time or compared to historical values for context. Orbital Insight’s GO Platform object detection feature is useful for generating data in wide areas — such as Yemen, that are both difficult to reach due to war and conflict and to capture data at scale.
Evaluating the Efficacy of Policies and Investment in Portland
Decision Makers also use the GO Platform to support policy and investment decisions. Using the GO Platform’s wide area and object detection features, an analyst created a dataset of all cars within Portland’s metropolitan zip codes to understand how gasoline tax increases have impacted total vehicles on the road since the tax was implemented in 2017. With GO’s open order settings, the analyst can watch progress on this policy issue and monitor its ramifications until the tax increase expires in 2020.
This same analysis also supports investigating the impact of Portland’s investments in both public transportation and bicycle lanes. Because these city investments took place during the same time period as the gasoline tax, the analyst was able to aggregate both datasets across the entire metropolitan area in addition to analyzing results at the zip-code level. As a result, the analyst was able to evaluate the dataset at both hyperlocal and city-wide scales.
Understanding Chinese Investment in Addis Ababa
Wide area traffic analysis provides useful data on congestion or air pollution in major cities. These vehicular detections can also reveal a picture of large scale infrastructure investments. Certain neighborhoods in Addis Ababa have been the focus on Chinese investment in recent years. The GO Platform’s vehicle detections across wide areas of the city demonstrate the impact these investments have made to vehicular traffic.
Whatever the problem set, we’ve designed the GO Platform to make geospatial data generation and consumption accessible to a broad range of diverse users with varying technical needs. GO users can access results of their wide area data creation in our User Interface (UI) or via API. Whether you choose to play a time series in our UI or export your results via API to combine them with other existing datasets, the GO Platform facilitates seamless integration into multiple data visualization and processing environments.
Orbital Insight’s GO Platform addresses many of the challenges associated with generating sufficient data for a better understanding of historic and near real time activity. We developed the GO Platform to help Clients understand what is happening across wide areas of the Earth. Through object detection and activity-based intelligence, our GO Platform ultimately facilitates better policy, safety, environmental and investment decision making.
For more information on gaining activity-based insights in our GO Platform, please reach out to: email@example.com