Megh Computing’s PK Gupta joins the conversation to talk about video analytics delivery, customization and more.
Megh Computing is a fully customizable, cross-platform video analytics solution provider for real-time, actionable insights. The company was founded in 2017 and is based in Portland, Oregon with development offices in Bangalore, India.
Co-founder and CEO PK Gupta joins the conversation to talk about analytics delivery, customization and more.
As technology keeps moving to the forefront with video analytics and smart sensors, what are the tradeoffs versus cloud deployment?
GUPTA: The demand for edge analytics is increasing rapidly with the explosion of streaming data from sensors, cameras and other sources. Of these, video remains the dominant data source with over a billion cameras deployed worldwide. Companies want to use analytics to extract information from these data streams to create business value.
Increasingly, most of this processing is performed at the edge near the data source. Moving the data to the cloud for processing incurs transmission costs, potentially increases security risks, and introduces latency in response time. Therefore intelligent video analysis [IVA] moves to the edge.
Many end users are concerned about sending video data externally; What are the options for on-premises processing and cloud benefits?
GUPTA: Many IVA solutions force users to choose between deploying their solution on-premises at the edge or hosting it in the cloud. Hybrid models allow on-premises deployments to take advantage of the scalability and flexibility of cloud computing. In this model, the video processing pipeline is split between local processing and cloud processing.
In a simple implementation, only the metadata is forwarded to the cloud for storage and search. In another implementation, data ingestion and transformation occurs at the edge. Only frames with activity are forwarded to the cloud for processing for analysis. This model is a good compromise between latency and cost between edge processing and cloud computing.
Image-based video analytics has historically required filtering services due to false alarms; How does deep learning reduce them?
GUPTA: Traditional IVA attempts have failed to meet enterprise expectations due to limited functionality and poor accuracy. These solutions use image-based video analytics with computer vision processing for object detection and classification. These techniques are error-prone, which necessitates the provision of filtering services.
In contrast, techniques that use optimized deep learning models trained to recognize people or objects, coupled with analysis libraries for the business rules, can essentially eliminate false positives. Special deep learning models can be built for custom use cases like PSA compliance, collision avoidance, etc.
We often hear “custom use cases” with video AI; what does that mean?
GUPTA: Most use cases must be customized to meet the functional and performance requirements for deploying IVA. The first universally required level of customization includes the ability to configure the surveillance zones in the camera’s field of view, set analysis thresholds, configure the alarms, and set the frequency and recipients of notifications. These configuration options should be provided through a dashboard with graphical interfaces to allow users to set up the analytics for proper operation.
The second level of customization involves updating the video analytics pipeline with new deep learning models or new analytics libraries to improve performance. The third level involves training and deploying new deep learning models to implement new use cases, e.g. B. a model for detecting PPE for worker safety or counting inventory items in a retail store.
Can smart sensors like lidar, presence detection, radar, etc. be integrated into an analytics platform?
GUPTA: IVA typically only processes video data from cameras and provides insights based on analysis of the images. And sensor data is typically analyzed by separate systems to glean insights from lidar, radar, and other sensors. A human operator is inserted into the loop to combine the results from the different platforms to reduce false positives for specific use cases like tailgating, employee authentication, etc.
An IVA platform that can ingest data from cameras and sensors through the same pipeline and use contextual analysis powered by machine learning can provide insights for these and other use cases. The contextual analysis component can be configured with simple rules and can then learn to improve the rules over time to provide highly accurate and meaningful insights.