Drone image processing

Aerial image processing
drone object recognition revealed boat on forest river

In recent years, Unmanned Aerial Vehicles (UAVs) have been adopted to provide services in numerous applications. Due to their ability to cover wide areas and capturing ultra-high-resolution aerial images at low cost, drones became an extremely important source of visual information.

At the moment the amount of information that comes from drones started exceeding the capacity of manual processing and automated object recognition systems came up in first place. There’re two approaches for implementing intelligent computer vision into UAV-based solutions: running recognition on board a flying drone or process recorded footage post-factum on a dedicated server or in a cloud. As an example for the latter case: drone flies along the route, capturing video from the area. After it has landed, personnel don’t need to spend hours to watch carefully the footage. Instead, all the data will be downloaded and quickly processed by the automated recognition system. The personnel will get a right to the ready-to-analyze report of detected objects/anomalies, statistics, etc.

drone image processing detecting a car

BitRefine offers a powerful recognition solution that is capable of processing high-resolution aerial images, made by drones, revealing standard objects, such as people, vehicles, houses, as well as custom project-specific objects, such as defects on various surfaces. BitRefine Heads multi-purpose recognition platform is based on a configurable video processing pipeline. Its first element is video acquisition that supports video files, sets of images, video streams from IP-cameras and machine vision GenICam cameras. After video source user adds interchangeable neural modules according to his task. If he needs to collect information about people or cars, he adds a corresponding neural recognition module from a library of available modules. This neural network will detect, locate and classify target object and pass information to the next stem along the processing pipeline. The user can add to the pipeline tracker, counter, color detector, and other elements to collect the required data and enrich it with details. With a flexible modular recognition pipeline, the user gains a significant level of freedom in designing a recognition solution that will solve his particular tasks in the most effective way.

In addition, BitRefine offers service of preparing customized neural modules that detects very specific objects, defects or visual patterns. Our specialists select the most suitable neural architecture from our module library and re-train it with sample images of the target object, provided by the client. After a neural module is trained, it can be added to the aerial image processing pipeline similar to all other BitRefine’s recognizers.

aerial image processing report

A video that goes through drone image processing pipeline ultimately gets converted into structured data, that contains objects, their locations, properties, and attributes. All this data is saved automatically into DB along with corresponding reference images. Recognition platform has built-in reports section that allows a user to see data, collected by UAV in the form of interactive charts and tables. If a drone collects statistics, there will be several found objects separately for each object class. If the drone’s task was a quality inspection of the remote facility, a user will get a table of detected defects together with their images.

Related Solutions

We’ve got more video analytics to share. Let’s be friends. But you first...