Planet Object Detection Platform: deep-learning object recognition as the new standard
Ivo de Liefde
Esri Certified Professional
Imagine: a GIS solution that recognizes your specific objects and quickly provides you with relevant, usable geo-data. It is a tool with which you can easily analyse recognized objects so that you save on costs and capacity.
The Planet Object Detection Viewer is that solution and can be used now. The tool automatically recognizes diverse physical objects based on visual material such as aerial photos, satellite images and video. This innovation heralds a new chapter in the field of object detection.
In this article, I will discuss the function of the Planet Object Detection Viewer and explain why this tool is so valuable within the GIS landscape.
The POD viewer is a deep-learning geographical information system (GIS) that works on the basis of a trained algorithm. The results? New, faster and more accurate insights from object detection.
Want to see everything about the Planet Object Detection Platform and its possibilities in one overview? Then download the Planet Object Detection Product Sheet.
Growing demand for Object Detection
We were increasingly being asked by the market about whether we could recognize specific objects on the basis of visual material such as aerial photos or video. As a GIS specialist, we could of course help with that, but we did not have a clear system with which we could define and analyse multiple types of objects in a relatively short time.
That was the reason for developing a tool that itself recognizes and analyses objects. We succeeded with the Planet Object Detection Viewer! And I do not mean it is just a handy tool. No, the POD viewer can itself learn, recognize, analyse and learn more based on AI. In other words, machine learning in object detection.
The variety of objects and types of visual material to be used makes the solution unique and thus very valuable for organizations.
By properly mapping the data for objects, you as an organization can better streamline processes and respond more effectively to needs, changes or problems. The process of collecting, enriching, visualizing and analysing geographical data is quite a technical and, above all, expensive and time-consuming process.
As a result, we see the growing need to accelerate the geo-data process (generation, processing and analysis) from object recognition. Below is a selection of possible applications:
A municipality wants to know where and how many dormer windows have been placed within a certain area. And it wants to check this periodically to find out whether the number of dormer windows is synchronous with the number of permits issued. The same process is conceivable in the area of tree felling and planting, expansion and extension, et cetera;
The fire brigade for a specific region wants to know which buildings have a thatched roof. The reason for this question is to inventory the risks of fire. We received a similar question from an insurer, but then focused on chimneys;
A government wants to map the number of solar panels on houses to find out how sustainable a region is and what the development has been. Solar panel recognition can also be used for organizations active in the sale and placement;
A province wants to know which verges must be mown for a good overview of intersections and traffic signs. Object recognition based on video can offer a solution.
In principle, we can train the algorithm based on each photo/video on which an object can be distinguished from the environment. This makes the number of applications for object recognition numerous.
How it works
To give you a better idea of how the POD viewer works, I will describe the working process below. The first step is to define and specify the object to be recognized. Based on this definition, we will prepare a first dataset and optimize it for 'training' the algorithm. This definition is also used to test the function of the algorithm later.
Based on the object definition and the data set prepared, we will train the algorithm to recognize that object as precisely as possible. The trained algorithm is then prepared for use in a (web) application of your choice.
The POD viewer can now recognize and analyse the desired objects. After selecting the object, select your area of interest. When you then start an analysis, the polygon you have drawn is forwarded to our FME server. From there, the (aerial) photo of the selected area is retrieved and fed to the correct machine-learning GIS model. This model is the algorithm that was previously trained to recognize the relevant objects based on a data set.
The algorithm will then make a 'prediction' on the selected area. The result is converted to a vector file that is reloaded and displayed on the POD website. These results can, if desired, be presented by the user on an interactive map or downloaded as an Excel, Shapefile or PDF file.
Want to know more?
Go to our website for more information about the how and why of the Planet Object Detection Platform. Would you like to try the tool right away? Register quickly and without obligation on https://geoai.tensing.com for the free version and use many functions immediately. We would like to hear what you think and any questions you may still have.