Object Detection Analytics

Data-driven solutions that ensure company security in Covid-19’s “Phase 2”

Since the beginning of “Phase 2“, a period of coexisting with Covid-19, both scientists and governments have agreed that the keywords for this period are: “security” and “social distancing.”
Navigating this phase is not an easy challenge. In order to overcome its obstacles there is a need for security protocols. These include thermoscanners, sanitization of environments and, most importantly, the use of protective equipment (such as gloves and masks). However, these measures will not be effective if the risks linked to gatherings in public areas or workplaces are not monitored.

In this piece, we will highlight one of the most discussed topics today: Object Detection. We will explain what it is, our point of view and the solution we created – based on our experience in the industrial field – and its real-life application with AWS and CeMeDi.

Get back to work safely with Object Detection Analytics

The asset of the world’s most powerful Tech companies

Gartner identifies in the Image Recognition, to which the Object Detection is related, enormous potential for business, for an apparently simple but incredibly powerful reason: technologies of this type can identify objects, people, buildings, places, logos and anything else that is valuable to consumers and businesses.
In fact, smartphones and tablets equipped with cameras have boosted this technology mainly in industrial and consumer applications. For example, logos, cars, monuments, wine labels and book and album covers can be identified by consumers’ smartphones by simply using a mobile app that accesses image recognition software in the Cloud. If the benefits weren’t considerable, the world’s most powerful technology companies such as Amazon and Google wouldn’t have already widely adopted it to improve their products and services.

Object Detection Scenarios and Applications

The pandemic will accelerate the use of Object Detection, i.e. the point where an algorithm detects and locates objects in an image. As the MIT Technology Review observes, this is already happening in many different areas other than the workplace to safely map spaces.
In Retail, for example, Image Recognition and Object Detection technologies help brands with standardize store controls and to achieve consistent results from all sales channels, enabling them to take business decisions based directly on shelf data.
Object Detection is also crucial for self-driving cars, since it allows them to recognize a stop signal or to distinguish a pedestrian from a streetlight.
Lastly, these technologies have reached a good degree of maturity in Healthcare. They can be used to automatically detect bone fractures, cell abnormalities and much more; for example to speed up the evaluation of possible pathologies, as shown by a recent study from Harvard Universitythat traces, thanks to neural networks, Covid-19 from X-ray images.

Security and Privacy

In a World totally disrupted by Covid-19, new tracking systems, facial recognition and Object Detection technologies are now studied globally to understand how the virus spreads and how risky situations can be avoided.
Privacy experts, though, fear that in the rush to implement these features key issues related to data collection, storage and user consent could dangerously end up being pushed aside. It, however, doesn’t have to be.
This delicate issue is also being addressed by giants such as Apple and Google which, as reported by Il Sole 24Ore, have made themselves available to implement tracing applications, announcing a series of changes that will be included in the API to track virus positives in line with the requests of the European Commission.
As they have been required to point out, we will need the consent of users, who will be able to choose whether or not to activate this solution: in addition, neither will our location be recorded, nor will data sharing with other users (or with Apple and Google) be allowed, in addition to the total guarantee that the data flow will be made more secure thanks to encryption via Bluetooth.  

Similarly, our Object Detection Analytics (ODA) solution does not store images: it only collects, analyses, and securely reports data.

In fact, any control system, from tracking to facial or object recognition, can become an acceptable and widely used method only if the individual’s privacy rights are fully protected; ensuring that data is collected, stored and managed responsibly and their use limited to the purpose for which it was originally adopted.

Iconsulting Object Detection Analytics solution

The goal is to help companies to safely manage the Covid-19 emergency through the use of a data-driven solution, which captures data from images in order to provide insight for both preventive and executive operations.
The idea originated from our experience in manufacturing, and identifies Analytics as the key asset to support the share of insights: both for commercial purposes – as an answer to how a new product will perform based on characteristics and historical sales of “similar” products, and for the detection of any product non-conformities originated on the production lines (or potential blocks in the production line caused, for example, by “groupings” of products).

Thanks to these skills acquired in the field, we have developed a solution able to offer  real support to companies and enterprises that nowadays are called to manage a difficult coexistence with the covid-19 virus and, at the same time, ensure greater security. While fully observing privacy protection, our data-driven approach focuses on data and its analytical approach to guide organizations in ensuring a truly safe relaunch into their social and work environment.

Our solution enables you to:

• Detect the absence of personal protective equipment (mask)
• Detect the correct distance between individuals;
• Detect any gathering;
• Manage access control and queue management;
• Provide a statistical analysis dashboard with trends and coverage of occupied spaces.

Solution approach

The proposed solution is modular and highly customizable according to needs. In addition, there are many technologies that can be used and it is therefore possible to adapt the entire architecture according to the application context.  

The main modules required to operate the solution are:  

• Image capture system;
• Temporary storage for images (not foreseen in edge solutions);
• Artificial Intelligence system;
• Data storage and processing system for the input of the Image Recognition process output data;
• Output management system (Data Visualization, IoT devices, alerting systems, etc.).

In order to define a tailor-made solution and adapt it to the application context, a workflow is followed to identify the main requirements:

Image Acquisition System Identification
The image acquisition system is the main data source of the solution and needs the correct set-up to allow an efficient operation of the whole architecture. Both devices already present and used for other purposes (e.g. video surveillance cameras or thermal imaging cameras for body temperature identification) and new devices can be used. The main aspects to evaluate are:

• In case of already installed devices, the compatibility for digital image acquisition must be evaluated;
• Positioning and framing of the device;
• Frame-rate image acquisition.

Calibration
Identification of the parameters and operating variables of the image acquisition device in order to allow a tuning of the algorithms that deal with the image recognition process (fundamental for the calculation of distances)

Space analysis
The environments analyzed through the Object Detection algorithm must be carefully studied to optimize the results obtained by the solution and to calibrate the algorithms used. Through a survey of experts we will be able to identify the most correct positions where to install the image acquisition devices (identification of elements that cause image occlusion, blind spots, etc.)

Architectural design
The architecture can be implemented in both on-premises and Cloud environments aligning with the needs, policies or systems already present in the company’s technology stack. Based on these parameters it will be possible to design the final architecture, with particular attention to limit the production time of the solution and respect for privacy. The architecture will be designed in collaboration with the main corporate reference figures, such as:

• IT Managers
• Security managers and DPOs
• Legal Managers
• Logistics Managers
• Human Resources Managers

Prototype and testing
Based on the requirements identified, a working prototype will be built on which the testing of Image Recognition algorithms and models in a controlled environment will be performed. Prototype and tests will be useful to perform a fine tuning of the solution and parameterization of the system.

Data Visualization
The tools to be used for the visualization of the analysis on the data generated through the image recognition solution will be defined, evaluating the possible reporting and Data Visualization technologies already used in the target company context. Specific KPIs and the best visualizations to represent the analyzed data will be defined and implemented. In this phase will also be defined workflows of notification and alerting.

Solution industrialization
Once the application context and the final parameters have been identified through the realization and testing of the prototype, it will be possible to industrialize the final solution for production through batch processes and streaming.

Internet of Things
One of the application contexts of the solution is to be able to control IoT devices through the data output obtained from image recognition processes. In this context streaming environments, queue management and development of communication processes with target devices will be defined.

---