Exclusive Talk with Patrick Elliot, CEO and CoFounder of VisualCortex

Q1: Tell us about how the computer vision space has evolved and your journey within it.

Patrick: Interestingly, the computer vision movement really began back in 1969 when New York City installed cameras on its municipal building. Most of the analysis was manual, but the concept of collecting data from video had begun. In fact, computer vision used to be referred to as VCA – Video Content Analysis. Today, we are actually in the fourth generation of computer vision. In essence, computer vision – or video analytics – is the ability to use machine learning models to perform automatic detections and data collection from within video footage.

Regarding modern use cases to which today’s video analytics technology can be applied, there are many possibilities – such as counting people or reading license plates. Whilst VisualCortex’s Video Intelligence Platform can facilitate any number of different computer vision use cases, internally, we stringently focus on training ML models that are globally repeatable and will provide real-world outcomes. We don’t chase every fire truck, so to speak. 

It is important to note that there are many Proof-of-Concepts in the industry or research being conducted with limited scope or depth. This is not our approach. We only work on video analytics use cases with enterprise outcomes and ML models that produce very high accuracy and confidence levels.

Q2: How does VisualCortex’s video intelligence platform connect computer vision’s potential to real-world business outcomes? 

Patrick:  VisualCortex is connecting the promise of computer vision and machine learning to real-world business cases by bringing an enterprise-ready approach to video analytics. It’s making this happen in four ways:

• Providing a lower barrier to entry by enabling organizations to use their existing infrastructure and commodity hardware
• Delivering plug-and-play access to quality-controlled models: The VisualCortex Model Store makes it easy to get started quickly with out-of-the-box, third-party, and BYO machine learning models

• Ease-of-use drives short time-to-value by enabling non-technical people to understand and act on insights derived from video analytics
• A robust platform approach, which:
  • Provides the stability to productionize machine learning models and deploy them throughout the enterprise
  • VisualCortex is also adaptable to new use cases from across different industries. It is not a point solution that’s locked into focusing on one vertical, business area, or outcome.

Q3: What are some of the biggest challenges in the video intelligence field? 

Patrick: Firstly, a lot of people are too ambitious on day one. You should focus on the most accessible use cases, which can deliver the most value, in the least time, and progress from there. From a technology standpoint, we see organizations most often struggling with the following: 

• Camera side or point solutions – which lack the scalability and flexibility to apply to meaningful, and multiple, business challenges 
• Tools that are hard to deploy commercially – financially and technologically
• Technology that’s aimed at data scientists, who aren’t business decision-makers

Q4: What are some ways that customer data can be acted on?

Patrick: Whilst we are under NDA, we can give you some good examples, which ostensibly pertain to people and vehicle detection. We’ve already completed a transport-related Proof-of-Concept with Servian, which analyzed road network utilization for a government client. This included automating things like vehicle counting, traffic pattern analysis, and anomaly detection.  

We are also working with well-known commercial real estate conglomerates to improve things like car park management. This can include things like: 

• Lowering ongoing operational costs by streamlining entry and exit protocols with automated License Plate Recognition (Our current version of LPR in car parks is yielding greater than 95% accuracy)
• Increasing revenue by improving the percentage of accurate license plate reads
• Enhancing customer experience by reducing wait times and eliminating error-based delays

In the retail space, we’re helping a nationally recognized brand to: 

• Track foot traffic effectively, efficiently, and repeatably
• Optimize in-store merchandising by analyzing shopper dwell-times
• Optimize staff rostering and reduce customer service delays;
• Capture ideal customer service examples for staff training purposes.

The more our dialogue progresses, the more use cases are emerging.

We’re also embarking on a Smart Cities initiative, which will be globally repeatable and is a natural extension of the work we do with people and transportation.

Q5: How does VisualCortex differ from its competitors? 

Patrick: Unlike camera-side or point solutions – which typically focus on one video analytics challenge per deployment – VisualCortex delivers a highly performant enterprise-grade platform to facilitate any real-time or historical video analytics use case. Because it was designed for enterprise deployment from the ground-up – rather than a result of a speculative Proof-of-Concept – it’s easy to deploy, use and add new use cases as they emerge. 

Clients are able to leverage commodity hardware to perform video analytics at scale. Many video analytics solutions rely on in-camera technology or specific types of cameras, thus forcing new hardware purchases and significantly adding to the Total Cost of Ownership. 

Through the platform approach, we can readily adapt to changing client needs. Once we’ve started a video analytics project, we often find that three or four other use cases quickly emerge. With VisualCortex, you can securely and efficiently run multiple production-level machine-learning models on each of your video sources.

VisualCortex can also be deployed anywhere clients desire – from on-premise and on-the-edge to public and private clouds, or a hybrid approach. This significantly reduces the barrier-to-entry for our customers and means they can get started faster.

Q6: How does VisualCortex store, analyze, and act on all customer data?  How does VisualCortex protect the privacy and security of its customers?

Patrick: The question of data security and privacy is one we think about a lot. Firstly, no matter where we process video content, we do not store that content.

Once we’ve processed the video in question, we discard the footage itself. What we keep is the video metadata, which is what we use to fuel use cases – to identify defined objects and actions as they occur within the original video source files. 

In terms of data processing, we can do that on-the-edge or in the cloud. The client can decide what makes sense for them from an efficiency and security perspective.

As with data processing, we have have a range of deployment options: 

1. On our cloud, as a fully-managed service; 
2. The clients’ cloud; 
3. Public cloud; 
4. On-premise and at the edge; 
5. Or hybrid

It’s completely up to the customer.

Q7: What advice do you have for organizations looking to harness video analytics technology? What critical success factors do you recommend they consider?

Patrick: There’s four core pieces of advice I can offer here: 

• Keep your project scope tight and don’t let it creep 
• Align chosen computer vision use cases to business value – how will this enhance operational efficiency, increase revenue or improve strategic decision-making?
• Establish a pragmatic rollout strategy: Tackle the most accessible and impactful use cases first to achieve early success and demonstrate value to executive sponsors
• Engage your stakeholders throughout each phase to ensure the insights being produced are as useful as hoped and are actually being used

Q8: What do you foresee as the biggest trends in computer vision and video intelligence in 2023?

Patrick: Today, the video analytics market is highly fragmented. Specific solutions are being produced by consultancies to help deliver Proof-of-Concepts. Hardware manufacturers and surveillance companies are also developing in-camera AI capabilities.

As more organizations look to harness video content for analytics and decision-making, the mushrooming number of ad-hoc video analytics solutions will become problematic. Organizations don’t want the expense and hassle of purchasing and deploying new solutions and cameras to meet each computer vision use case. 

Increasingly, organizations will need a single platform – capable of running multiple ML models across a multitude of existing video sources – to produce analytical insights about all their video content in one place. From people counting and dwell-time analysis to determine customer exposure and engagement in retail settings; to vehicle detection and tracking to better understand road usage.