Videoloft’s recently launched analytics features are based on neural networks. Object detection and text recognition technology allows users to search their video for specific items and text. But we’re not stopping there… we’ve got another neural network model in development, we’ve named her Netty and we need your help training her!
Neural networks are the basis of deep learning, which is a subset of machine learning. Neural networks, especially those used in computer vision, are based on the structures of the human brain that deal with vision and interprets images. By mimicking the way a human brain works, they can read information, learn from it and then predict the outputs of other similar information. Once neural networks have completed training they enter inference mode. In this mode, they’re no longer learning but instead showing the capabilities that they have previously learned.
In the case of our newest neural network, we’re teaching Netty to read images, learn to recognize them, and thus understand the next time she sees such an image. The more images she sees, the more accurate she will become.
Neural Networks within the Security Industry
The accuracy and usability of neural networks has dramatically improved over the last few years and have been used in numerous fields.
In 2017 Kaggle, an online community of machine learning and data scientists, ran an online competition to see if someone could create one that could spot lung cancer. The result was phenomenal – the neural network which was created could read CT scans better than qualified specialists in healthcare. This project sparked the usage of neural networks more broadly in the healthcare industry. In airport security, neural networks are being built to detect concealed items on passengers from the body scans. What could this mean for the security industry and the CCTV industry more generally? The possibilities are endless! One interesting research project examines whether neural networks can detect violent behaviour in the video footage that it ‘sees’.
Although most recent dramatic improvements of AI have been based on neural networks, much of the security industry and its computer vision algorithms haven’t used them for most tasks. Why? Because only recently have they become efficient and accessible enough for a wide number of people to be skilled at training and deploying them. In addition to this, most CCTV footage is stored and processed at the edge. Simple DVRs and NVRs often don’t have the capability to run the AI applications. Given the greater efficiency and the further use of neural networks, over the next few years, we are likely to witness exponential improvements in their capabilities.
Right now we are only at the very beginning of their potential. We’ve seen lots of incremental technological improvements in the security industry, but in order to truly advance past our current capabilities neural network adoption will be vital.
Why are Videoloft creating neural networks?
Firstly, we’ve been quite lonely over the Covid-19 lockdown period and we would like some new friends. But more importantly, we want to bring the best analytics tools to our cloud video platform. While more common artificial intelligence tools achieve simple analytics tasks, they don’t perform very well when trying to identify whether a spider web across a camera lens is a viable motion event or a false positive. So this is Netty’s first challenge – to read thousands of images from cameras covered in spider webs so that she can reliably recognize them in the future.
Once Netty completes her strict training programme, Videoloft will be able to alert users who have spider webs causing false positives. And in future versions we hope to be able to remove the false positives altogether. We’ll continually improve the Videoloft software capabilities adding new established neural networks modules via remote software updates. Videoloft customers won’t need to buy expensive hardware to get impressive functionality, that’s the beauty of intelligent software! Videoloft aims to make neural networks accessible for the whole security industry. But we need you to help us get started.
We need your help!
We want to make Netty as powerful as she can possibly be. To do that we need volunteers to help train her! We’re asking our global network of security professionals to give us access to an outdoor camera covered in spider webs. The more volunteers we have, the quicker she’ll learn. If you want to give Netty a helping hand, email us at partners@videoloft.com and we’ll give you more details. To say thank you, we’ll send you a free Cloud Adapter to use on your next cloud installation.