Deep Insights with AI for Video Analytics
There has been a revolution brewing in the technology industry. An artificial intelligence (AI) method called deep learning that uses deep or multi-layer neural networks is dramatically improving computer vision and video analytics. So much so that deep learning-based computer vision can now beat human capability in rapidly identifying objects in images.
With video technology all around us — it’s this deep learning that can help us process vast amount of data that has been too great to humans alone to process. Now, we at IBM are taking this technology to the next step.
Today we are announcing the integration of PowerAI Vision and IBM Intelligent Video Analytics, allowing organizations to build customized AI models for detecting objects or activities of interest that can be easily integrated into IVA for easy model deployment.
Applying Deep Learning to Real-Time Video
By leveraging powerful algorithms and deep learning technology, Intelligent Video Analytics (IVA) analyzes live video streams to quickly identify and examine potential issues, anomalies, or particular events of interest. Designed to help commercial and public organizations better leverage video to enhance infrastructure efficiency and safety, IVA automatically transforms raw video into enriched, searchable, structured data. IVA can help users to quickly highlight potential subjects and objects of interest, detect or track patterns, observe heat maps and anomalies in video footage, and deliver rapid redaction of subjects or objects.
PowerAI Vision: Point-and-Click AI for Video and Images
Training AI models using deep learning is an arduous task that requires preparing and labeling the data, electing the appropriate neural network model, and choosing and tuning hyper-parameters during training to improve AI model accuracy.
The PowerAI Vision software suite uses Auto-ML (automatic machine learning) methods to automatically train deep learning models for object detection and image classification. PowerAI Vision takes a data set of images or video frames labelled with the objects to detect, selects an appropriate neural network model, and trains the model by picking the right hyper-parameters for best AI model accuracy. The software accelerates prototyping by enabling non-data scientists to train AI models for images and video. See the video demo below.
Unlocking Business Value from Video with AI
The new integration of PowerAI Vision with IBM Intelligent Video Analytics (IVA) means that we can now create new AI models for recognizing specific objects and integrate these models into IVA for further analytics and action as shown in the figure below.
By auto-training these new AI models in PowerAI Vision, we can now use IBM Intelligent Video Analytics across a variety from industries, such as:
Remote Inspection and Asset management — Power companies are transforming inspection jobs by deploying drones with cameras to capture data for inspection. Clients can use PowerAI Vision to build customized AI models to automatically flag equipment that might be worn out, in need of inspection, or likely to fail. These custom models can be integrated into IBM IVA to automatically schedule maintenance jobs.
Retail — Real time video analytics can help retailers in a variety of ways, including the generation of enhanced customer insight, inventory management tools that signal robots to stock shelves, and AI-powered cameras that can catch fraudulent actions at the point-of-sale. Custom models for each of these built in PowerAI Vision can be integrated with IBM IVA to greatly enhance the shopping experience.
Worker Safety Compliance — We can create a custom vision AI model in PowerAI Vision that can identify safety helmets, jackets, gloves, et cetera. IVA enables setting up different safety zones and use different AI models to flag worker safety compliance per the zone requirements.
Video represents a hugely untapped resource for data-driven insights, and now with this exciting integration Intelligent Video Analytics with PowerAI, IBM is helping unlock its potential. The integrated offering will be available late Q2 2018.
Attend our upcoming webinar on PowerAI Vision to learn more about that software. I would love to hear your thoughts on how you can use these kind of capabilities together.