Alternative Use of Imaging Systems through Artificial Intelligence (ALTRUIS)

Filter items

Initiator(s):

Digipolis

Theme:

Digitalisation

Domain:

Governance

Active state:

Realised

Contact:

Digipolis
Christophe Marrecau
T +32 473 94 50 77

Ann Fournier
T +32 478 90 14 59

PIP
Erwin Dewallef
T +32 472 72 40 48

Tabs

Challenge

With the PIP project Alternative Use of Imaging Systems through Artificial Intelligence (ALTRUIS), Digipolis wanted to contribute to handling the mobility issues in and around the city of Antwerp by using camera images in an alternative way and by applying Artificial Intelligence (AI).

Every day the city of Antwerp is confronted with one of the heaviest traffic congestions in Europe. To remedy this situation, the city's major mobility axes are currently being thoroughly reorganised. The accompanying construction works (also known as 'the construction site of the century') put even greater pressure on the mobility and the quality of life in and around the city. A disruptive approach is therefore needed to meet these challenges.

Digipolis wants to contribute to this by firstly applying innovative techniques such as AI in the Smart Zone. This is the demarcated central part of the city of Antwerp which has been designated as an experimental zone for Internet of Things (IoT) and is therefore equipped with a wide range of devices such as sensors and cameras.

Within the framework of ALTRUIS, Digipolis wanted to use the visual footage from the cameras, which have initially been installed for law enforcement purposes, in an ‘alternative’ way and beyond the law enforcement area with an eye to sustainable mobility solutions.

To this end, with ALTRUIS, Digipolis envisaged the development of a disruptive image recognition system that, supported by AI (Deep Learning Neural Networks), would be able to interpret the street images in such a way that traffic can be mapped in various manners: counts, classification of mobility vectors (cars, buses, trams, [motor]bikes, pedestrians, etc.), speed, sense and direction of each road user, detection of the trajectory covered by mobility vectors and of anomalies (congestion, conflict situations, road accidents, etc.), etc.

This extensive knowledge of mobility should allow for a smart(er) reorganisation of the mobility (firstly) in this zone, both through immediate feedback to the city, its residents and visitors (dashboards, smart signalling, apps, etc.) and through more structural measures (adaptation of the traffic infrastructure to new insights, alteration of [local] traffic rules, control of smart traffic lights, promotion or discouragement of specific mobility vectors, sustainable organisation of [local] transport and deliveries, etc.).

In short, ALTRUIS aimed at upgrading through interpreting street camera footage, which in itself has little value, to mobility context and insights with an eye to smart mobility solutions in the city of Antwerp (firstly in the Smart Zone).

Preparation

Preliminary research

Digipolis has carried out its own preliminary research to see what is already technologically possible for the intended disruptive image recognition system (e.g. [open source] algorithms to convert the camera images into data) and what still needs to be developed. This preliminary research was completed with a meetup on 18 October 2018 where a number of interesting speakers spoke about image recognition. More information about this meetup is available here (information in Dutch only).

With the insights gained from the preliminary research, Digipolis decided to tender the realisation of the ALTRUIS image recognition system in the following phased way:

Anonymisation of image material;
Image recognition and algorithms for insights;
Testing and validation of working AI prototypes.

Procurement

Digipolis has put out to tender phase 1 (anonymisation of image material) and phase 2 (image recognition and algorithms for insights) of ALTRUIS combined in a single contract, shifting the technological scope from 'Graphics Processing Unit (GPU)-based AI (in the cloud or on premise)' to 'AI at the edge (edge AI)'. In concrete terms, this means that, instead of bringing sensor data to the cloud and processing it with AI (which is insufficient for systems that need to be able to decide/interact in milliseconds), AI is brought to the sensors in order to make the Internet of Things (IoT) really 'smart'.

The contract Development of basic infrastructure for image recognition with edge AI tendered by Digipolis included the following:

Setting up a Proof Of Concept (POC) with the basic infrastructure for the analysis of (camera) images on edge devices;
The elaboration of two use cases: (1) (pseudo) anonymisation of images, and (2) object detection and classification applied to traffic measurements.

The contract did not yet include the development of hardware but merely formulating advice on the specifications of the required hardware (e.g. edge devices) and a proposal for the total (edge computing) architecture picture.

The contract was put out to tender via the tried and tested tender procedure Buy from start-ups of Digipolis. This is a 'light' tendering approach specifically aimed at involving small(er) innovative market players in contracts of Digipolis and combines the negotiated procedure without prior publication with some aspects of Pre-Commercial Procurement (PCP). Buy from start-ups is a two-step procedure in which first suitable candidates for a contract are selected and then negotiations are conducted with the selected candidates who have submitted an offer. These negotiations result, whether or not via a Best and Final Offer (BAFO), in the award of the contract.

The contract for the development of basic infrastructure for image recognition with edge AI was awarded to the supplier Rombit NV.

Results

The twofold contract (1) setting up a Proof Of Concept (POC) with the basic infrastructure for (camera) image analysis on edge devices and (2) working out two use cases was carried out and delivered by Rombit NV.

Digipolis has investigated the possibilities of a next phase ('testing and validation of working AI prototypes') of ALTRUIS, with in the first place the further maturation of the basic solution (POC) developed by Rombit NV and subsequently a possible upscaling to a production-ready solution.

However, due to an (internal) reorganisation and budget constraints, Digipolis has decided not to pursue a next phase of ALTRUIS. As a PIP project, ALTRUIS has therefore been completed. Port of Antwerp, with which Digipolis entered into a partnership within the framework of ALTRUIS, is continuing with ALTRUIS and is having the POC further developed by Rombit NV in order to obtain insights into the movements of shipping traffic. This continues to take place outside of the PIP.