About the client
Vexcel Imaging is a top-tier aerial photography company specializing in custom hardware and software solutions for map creation and image analysis.
Their technology empowers clients to access 2D and 3D spatial data of their premises and leverage visual data efficiently through advanced algorithms, including computer vision.
Project description
Vexcel Imaging partnered with Vention when their image analysis framework was still largely manual. The framework allowed them to stitch multiple images into collections for specific territories and identify imperfections or blind spots in imaging.
The challenge? The manual nature of the process consumed significant time and resources. It required extensive effort to account for image angles and perspectives, and it also limited their ability to detect dark spots — areas or even individual pixels not captured by cameras.
This is where our computer vision and drone software development expertise came into play. Vention’s mission was clear: to automate image processing, boost efficiency and speed, and reduce imperfections to deliver a seamless solution.
Our solution
The solution was built on the client’s AWS platform, leveraging AWS Step Functions and Lambda functions for maximum efficiency.
Image analysis pipeline
We tackled the challenge step by step, starting with the foundation: the image processing pipeline.
Vention’s team developed custom algorithms that scan, analyze, and gather metadata from images — which enabled Vexcel Imaging to quickly identify critical issues, like black pixels or corrupted files, and automatically exclude such pictures from their project pool. By automating this step, the team significantly reduced wasted time and effort on unusable images.
The pipeline also autonomously analyzes and modifies images, seamlessly combining them in the required order and format while accounting for differences in angles and perspectives. As a result, Vexcel Imaging specialists can dedicate their time to creating products and analyzing data rather than spending hours manually sorting through images.
AWS-based serverless automated solution
For projects like this, a serverless architecture often outperforms traditional server-based infrastructure for two main reasons:
- Cost-efficiency: With serverless, you pay only for what you use. When no images need processing, no Lambda functions are invoked, resulting in zero cost. Such flexibility is particularly beneficial for workflows with varying data loads.
-
Instant scalability: A sudden influx of a million files? No problem. The solution instantly triggers additional functions and scales down quickly, eliminating the need for extra hardware or long wait times.
The workflow
-
Initial filtering: The solution first analyzes the pictures, automatically excluding those with black pixels or corruption.
-
Format standardization: Next, it converts all images into a unified format to streamline data gathering, analysis, and further processing or modifications.
-
Image refinement: Mathematical models then analyze the images, identifying and comparing their borders. The system adjusts them through rotation or shearing, eliminating distortions and deviations to simplify processing.
-
Metadata compilation: Finally, the images selected for the final product undergo metadata extraction. This data is stored in a separate file and provides comprehensive information about the areas captured in the images.
Why Lambda Functions?
Lambda functions were the optimal choice for launching numerous parallel processes. A typical dataset from one area can span five to seven terabytes, and analyzing it sequentially, even with automation, would take days or longer.
With a 15-minute runtime limit but unlimited parallel instances, Lambda functions allowed us to break the workload into smaller image groups. By leveraging AWS infrastructure, we processed all images simultaneously, which significantly accelerated the entire workflow.
Key stats
years of active collaboration
Vention team members
Image processing now takes hours — not weeks
Results
When Vexcel Imaging implemented our solution, they gained the ability to automatically process up to 1,000 images in just 15–20 minutes (instead of weeks!), seamlessly handling datasets of over 10,000 images.
Beyond the speed boost, we significantly reduced manual effort, freeing their team to focus on high-value tasks instead of sorting through batches for unsuitable images.
Tech stack
Backend
Python
Rasterio
GDAL
Shapely
NumPy
Docker
Data storage and databases
AWS S3
PostgreSQL
PostGIS
Infrastructure
AWS Lambda
AWS API Gateway
Terraform
AWS Fargate
AWS SQS
Jenkins
AWS Step Functions
