A Halftime Review of Our 2019 Video Content Analytics Predictions
2019 Video Content Analytics Predictions vs. Reality
We’ve reached the halfway point of 2019, which is a great time to take a look at the predictions we made for this year in video surveillance and video content analytics (VCA) and evaluate how industry trends have evolved in light of those predictions. In December 2018, the top three trends we predicted three trends for 2019 were: 1) increased adoption of face recognition technologies, 2) standardization of artificial intelligence-driven video analytics and 3) increased migration to edge and cloud processing, respectively, for supporting intelligent video analysis. Below is an update on where those trends currently stand.
Wider Adoption of Facial Recognition Technologies
Facial recognition technology is indeed becoming more popular and more widely adopted, because it has become more accurate (and therefore more reliable) due to advances in technology and cameras that provide photos and video with higher resolution and quality. Made possible by Deep Learning and artificial intelligence (AI) technology, facial recognition matches or verifies people by correlating biometric features from digital sources, such as video frames.
The technology is readily available and is proliferating throughout the world; as an example, it is often used as a means of logging into a smartphone, or identifying friends in social media photos. When it comes to video analytics-driven face recognition, video surveillance operators can create a watchlist of uploaded images or screen captures to define which people or objects the system should detect and identify. Based on the watchlist imagery, automatic alerts can be triggered when matches are found in real-time surveillance. These powerful capabilities are significant for law enforcement and security organizations that need to respond rapidly to dynamic conditions and make informed decisions quickly. Police, for instance, use the technology when children or elderly adults are reported missing. The ability to search video and trigger alerts for appearances of missing people is critical for investigating and understanding the circumstances of the person’s disappearance and driving their swift recovery and return to safety. Similarly, retailers could use the technology to immediately identify past shoplifting offenders and prevent future theft.
The use cases around the technology are compelling and adoption will continue to rise – especially as regulations around the responsible and legal use of the technology are defined. As with any technology, the deployment and use of face recognition must ensure the proper balance between privacy, data protection, and security through transparency and proper guidelines. With clearer compliance, as well as further technology advances, adoption of face recognition and biometric technologies will grow, as will its utility.
The Standardization of Artificial Intelligence-driven Video Analytics
Deep Learning, which relies on Deep Neural Networks (DNNs) to analyze images and videos, is now the de facto standard for video content analytics technologies. DNNs power Machine Learning, which is the subset of AI that trains a machine how to learn. The development of cluster/cloud computing, more data storage, and Graphic Processing Units (GPUs) that have more computing power are all factors that have made it much easier to leverage DNNs. The increased coverage and cost-efficient processing allow systems to continuously process more video and aggregate metadata over time.
Modern video content analytics technology solutions use GPUs and Deep Learning to translate live or archived video into structured data, extracting rich metadata, which makes the data more accessible and actionable – deriving deeper insight from previously underutilized video. Beyond alerting functionality, Deep Learning-driven solutions make it possible to uncover quantifiable data and trends from video metadata, derive actionable insights for business intelligence, and make data-driven safety, security, and operational decisions.
Now that AI and Deep Learning are enabling video analysis and other critical applications, the industry is focused on increasing analytic accuracy and cost efficiency. The demands of video analytics software are still quite extensive; tagging data and training the DNNs requires manual tagging and large volumes of data, and deploying the technology requires substantial computing resources. The next phase of VCA development will focus on streamlining the requirements and total costs of ownership of VCA solutions.
For example, in terms of technology development, DNN training and tagging can be made more efficient with synthetic data generation. 3D rendering software can manufacture realistic data based on objects in actual images and video frames. The synthetic data is created algorithmically, and can then be used to train Machine Learning models. Fabricated data is much easier and less expensive to acquire than real data, and infinite amounts of data can be generated to enable Deep Learning and to drive AI-backed technologies. Furthermore, because the data is generated based on a specific data point that is already identified and classified, it’s not necessary to manually tag it; the artificial data is already labeled when it comes into existence.
Flexible Architectures and Hybrid Video Content Analytics Deployments
The combination of increased high resolution video footage, the proliferation of video cameras and Deep Learning-backed activities incurs high processing demands to enable the desired video content analysis speed and accuracy. Technology providers are coming up with flexible, cost-effective system architectures to support these demands and to meet the varying needs of the end users. Whereas the need to conserve bandwidth is driving a surge in on-camera analytics and edge computing, cybersecurity advances are encouraging migration to the cloud, and small form factor hardware is emerging to support centralized computing of video data. These trends are not limited only to video analytics, but are now having an impact on video content analysis software.
The flexibility to support comprehensive VCA activities on the edge and in the cloud—in addition to traditional centralized processing models— is helping organizations adopt video intelligence software and achieve their analytic objectives at the lowest possible cost. Each model has its functionality benefits, and, today, a combination of processing models is optimal for most video surveillance implementations.
The Video Content Analytics Forecast
Other new trends have also emerged in 2019 and will likely continue to develop as the year continues. For example, we are seeing a movement towards comprehensive video content analysis platforms rather than point solutions. Organizations are seeking to adopt a complete product so that integration is seamless and not ongoing: By integrating a single product with comprehensive value, they seek to maximize their technology investment. With today’s advanced video analytics offerings, many companies are finding that this single technology implementation can be validated by multiple business groups within the organization beyond security and public safety.
The VCA space is evolving and it’s a particularly exciting time to be a part of the innovation and development on this front. Looking back at our earlier predictions, we can see how these trends have shaped the future of the VCA industry and will continue to do so.