Through constant research and development, AVUTEC evolved from exclusively recognising number plates to recognising all kinds of objects. With our knowledge and tools in video content analysis and image classification or detection, AVUTEC offers deep learning consultancy to develop any intelligent solution.
Video content analysis and data mining
Deep learning, machine learning and computer vision technologies are conquering the world by storm. We can now heat up our house from a distance, are introduced to new movies or music based on our previous choices and soon autonomous driving cars might be ordinary.
Since smart technologies are essential for companies to keep up in a world of constant progress, AVUTEC developed ready-made recognition modules and offers deep learning consultancy to develop custom video analysis routines, that understand and predict visual input. Besides video analysis AVUTEC offers data mining routines, that structure and sort out recognition data to trace insightfull patterns. Insights that can be used in data-driven business strategies, forecasting and decision making. These services fit any organisation, that has innovative ideas and wants to use the skills of a dedicated company.
The X-Series product line
Standard deep learning models require a vast amount of processing power. All AVUTEC neural networks and algorithms are optimised to work with embedded hardware. As a developer of software and hardware, AVUTEC products offer a complete AI solution for a successful deployment. The X-Series product line offers a wide range of AI camera systems, that use their internal neural processing power to achieve the best performance.
The combination of CortexFramework, a choice of (pre-)trained networks and the unique hardware constellation of the AI camera’s, offers products, that are unique in the world.
THE DEEP LEARNING PROCEDURE
AVUTEC’s deep learning service starts by defining the exact goals for the model. After targets are set the process consists of four consecutive steps: Data collection, data preparation, training and testing.
1. Data collection
The deep learning consultancy for video analysis always starts by defining clear goals. These goals are a starting point for collecting data and measuring the performance of the final model. Once the targets are set, the data collection can start. When data is collected, the quality of the data is a prediction for the quality of the results. The data needs to be representative of the actual circumstances the application will be used in. Poor data is equal to poor performance.
During the collecting and preparing of data, a strategy for the use of technology is made. This strategy defines the deep learning models and techniques that will be used, depending on the processing powers of the IoT device. Once the strategy is established, the actual training of a model can begin.
2. Data preparation
The next phase in the deep learning process is the preparation of the data. All visual material needs to have a predefined format with the same size and the same aspect ratio. Next to that, every image must be labelled, clean, consistent and accurate. Achieving this is a labor-intensive task, since it needs to be done with great precision.
Testing the trained model in the field is the final phase of the deep learning training trajactory. The application is released in beta version and a test protocol is applied, which results in generated reports, that display the performance of the trained model. As soon as targets are met, it is time to release the final version.
Read our deep learning articles
Watch our live detection demo
As an illustration of one of the base models, this live feed shows a real-time street viewing scene. The network is set to detect people, bicycles, cars, trucks, busses and more. The faces of detected people are blurred according to General Data Protection Regulations (GDPR). It serves as a sneak preview of all possibilities AI video processing has to offer.
See the live detection of CortexDetect.