Next generation
Building Inspection
To address the energy leakage from urban structures, The Inspectoire is taking the initiative to leverage thermal imaging as a non-invasive screening approach to identify and measure heat dissipation from buildings.
By combining the change in sensor data over time with clustering of similar buildings and normalized information about the local neighbourhood and analyzing it using artificial intelligence against the green standards, a solution is created to detect problems and provide early recommendations for the building owners and local or national administrative officers.
The solution combines the infrared thermal images and captured images of the building envelope and sends them to a cloud-based analytics pipeline. The pipeline components use machine learning techniques to reconstruct the structure by collecting overlapping photos based on the client’s position, camera angle, and distance from the structure, detecting anomalies and predicting future risks.
The attractive market for the product is the suburban and urban areas with aged buildings where energy conservation is a prime concern; both energy providers and consumers would be interested in evaluating and monitoring such structures to minimize the waste of resources. In Canada, Vancouver suburbs and Toronto have been identified as the starting points with the maximum possibility of traction for this product.
The business model, therefore, is to offer IaaS (Inspection as a Service), which will allow property owners to subscribe for periodic evaluation of their buildings, as well as the administrative city officials and power companies to combine this information with their current approach of sporadic inspection and smart metres. In that sense, an ‘enterprise-level’ offering has been devised to provide turn-key service for residential, commercial, industrial, and public property managers and inspectors, as well as contractors, builders, and insulation material providers.
We proposed a method to use segmentation in addition to semantic segmentation to enhance the performance of the semantic segmentation results. It is imperative the first time that the user will label the reference image. In this phase, we create a pipeline to extract super-pixel and segmentation for every image and combine the results with the semantic segmentation output.
To achieve this goal, for every segment in the segmentation, the method finds the semantic segmentation that has the most overlap with this segment and assigns this segment to the selected semantic class. This way, the procedure is able to preserve the border of the objects while matching the best labels to them.
Up to this point of the project, we worked mainly on extracting thermal information, semantic segmentation, and dataset generation. One other vital task to fulfill is image registration. In order to track the changes in the thermal pattern of a building, the user will take several photos from the same portion of the building with the potential of having energy leakage.
Therefore the application should be able to match two images from the same object at different times. The first image is referred to as a reference image in this work.
Not only the cooling and heating account for approximately half of all energy bills, but also it is an established fact that insulation leakage is responsible for most of the unexpected waste of energy in local communities. It thus follows logically that understanding energy leakage will help to reduce such unwanted emissions. Despite the progress in the development of new building management systems, energy waste remains high.
Therefore, thermal imaging has been introduced as a non-destructive inspection tool to audit insulation issues of building envelopes.
Continuous screening of buildings using AI-based modification detection will provide a reliable monitoring service to prevent unexpected energy leakages and allow for trustworthy recommendations on maintenance in the early stages.
Thermal imagers are, in essence, heat sensors that are capable of detecting tiny differences in temperature based on the infrared energy emitted by an object. This makes object detection and matching it to heat signature very critical. In addition, side objects such as shadows of neighbouring houses, obstacles in yards, moving objects in the picture, etc., can change the temperature footprint. To make matters worse, thermal imaging sensors are limited in terms of resolution and temperature difference.
Canada is a country with semi-arctic weather and a highly urbanized population, with more than 25% of houses constructed before 1960 and over 80 percent of its inhabitants concentrated in large and medium-sized cities. Based on the 2016 census profile, there are 14,072,080 occupied private dwellings in Canada, 25% of which were constructed before 1960 and a significant portion of them are classified as single detached or semi-detached houses, with this percentage growing with the age of the building.
These facts make Canada a receptive market with high demand for energy conservation, especially HVAC costs. On the other hand, the legal and tax implications of operating in Canada, the required licences and regulatory prerequisites, and the factors governing the valuation of ventures need to be considered in a careful market fit analysis to optimize marketing costs and create trust in the investor’s network by optimally managing cash flow and investments.