The innovation project was set as a proof of concept whose aim was to expand the tracking network of the cargo traffic in the port facilities, deploying the network of cameras already installed in the port area.
The challenge was achieved by the start-up AllRead MLT, which solution based on artificial intelligence is capable to spot and read in real-time any text and alphanumeric code from images or videos captured by any device (mobile, security camera, etc.) even when it is damaged, dirty or out of focus.
Unlike other market players, AllRead MLT´s solution is focused on the software, leaning on the benefits of Deep Learning, and not on hardware devices.
The scope of the proof of concept was the development and implementation of a minimum viable product (MVP) focused on processing the images captured by the existing cameras, in order to deliver the transit-related information in a structured way of the following items containers IDs, platforms, trucks, trailers and semi-trailers and all types of license plates of vehicles (mainly European and Moroccan origin).
The project plan was based on the following activities:
- Data Capture. Collection of images and videos exported from the APBA’s image capture and access control systems.
- Analysis and preparation of the data. Phase of selection, cleaning and labelling of the data collected for the training.
- Development of software / MVP to display the reading on a user interface and immediately record the data obtained, as well as the corresponding photo.
- Neural network training. Core” stage of the activity, focused on training and adaptations of the neural network models.
- Testing phase / Demo Day. Final demonstration of the solution and the results obtained in front of the business managers, using new videos not used in the training sessions.
The project’s innovation:
- Use of a convolutional neural network (a common Deep Learning technique) for the cargo tracking within the port area.
- Ability to identify, in the context of an image, the location and typology of a license plate (container, trailer, tractor, etc.), by understanding the framework and elements included in an image. This fact makes this solution unique and disruptive from the traditional OCR that reads characters without considering the context.
The project’s product:
- Software prototype to process the images captured by the existing cameras, in order to deliver the transit-related information in a structured way of the following items containers IDs, platforms, trucks, trailers and semi-trailers and all types of license plates of vehicles (mainly European and Moroccan origin).
- Increased accuracy rates in vehicle identification, reducing the number of false readings, and all this without the need to deploy new video and civil infrastructure or dedicated gates providing greater flexibility and coverage of the vehicle monitoring system.
- Extending the cargo tracking throughout the port area, monitoring not only light and heavy traffic vehicles but also the cargo transported by rail.
Next, a link to Youtube is suggested where a demonstration video of the developed prototype is shown: