Object Detection System Empowered with AI and Computer Vision with Learning Capabilities for New Objects
-
Industry
Manufacturing, Security, X-ray
-
Project type
Software modernization, Software development, AI, AI Transformation, DeepML, Computer Vision, DeepML, Model training, Softacom AI Lab
images used for model training
hours it takes for personnel to independently train the model
SDK based on GStreamer provided access to necessary video streams
successful tender was won by the Security Integrator
Description
In 2024, we were contacted by a technical security integrator company (let’s call it Security Integrator). This company planned to participate in a tender for the supply, installation, and personnel training for X-ray equipment. The equipment was designed to detect specific objects (primarily weapons) for a governmental facility (let’s call it Authority).
The Security Integrator is not a developer or manufacturer of X-ray detection systems. Instead, they are an official representative of a USA-based company that produces such equipment (let’s call it Manufacturer). The Manufacturer provides standardized software to clients globally, with no customization or tailoring to specific client needs. In our case, for Authority. Thus, the Security Integrator faced a challenge: how to meet the Authority’s requirements.
It should be noted that the Manufacturer’s software already included out-of-the-box functionality for detecting firearms in images and videos. But the Authority didn’t have an opportunity to add new objects into the system and train it to recognize those objects (for example, drones or explosives).
After the Manufacturer declined to customize its software, the Security Integrator, as our partner, turned to us to develop a solution.
- Adding new features to an existing proprietary system. Such limitations are often crucial as they influence the decision to obtain equipment and its compliance with tender requirements.
- Lack of ability to retrain the system based on the Authority’s data. This functionality is unique in implementation and use. Often, end users prefer pre-configured solutions without customization. But the Authority had the expertise and resources to contribute to the system’s development.
Solutions
After the Security Integrator contacted us, we conducted the primary business analysis and proposed a solution architecture. We introduced a separate AI- and DeepML-based detection system for augmentative learning. This system would operate independently of the existing X-ray software, displayed on a second monitor for the operator. The first monitor would continue to show the standard interface, while the new system would handle the enhanced detection functionality. This allowed us to bypass the limitations of altering the Manufacture’s software.
We also required access to the video streams from the X-ray equipment. After the negotiation process with the Manufacturer (the System Integrator, being a long-term partner, managed to arrange a fast communication process), we obtained an SDK based on GStreamer. As a result, we received access to necessary video streams.
Next, we defined objects to be detected, their detection probabilities, methods for how to inform an operator, and more. We decided to use a pre-trained machine-learning model as a base. This model was already trained to detect firearms (we’d used over 100,000 organic and synthetic images to train it). After validating the model’s performance (or neural network), we developed operator interface software.
A critical requirement was the system’s ability to undergo further training with the Authority’s datasets and expertise. This was achieved through Softacom AI Lab. We trained the personnel to work with Softacom AI Lab, explained how to prepare and validate data (photos), annotate data, and select suitable equipment for training the model within 4-6 hours using a dataset of up to 1,000 images per iteration.
Outcomes
- Thanks to the development of additional software functionality, the Security Integrator successfully won the tender.
- The Authority gained the capability to train the model and conduct its own experiments to detect required objects.
- Our company gained valuable experience working with X-ray systems and established a strong and collaborative relationship with the Manufacturer.
