Video analytics can benefit various verticals, including manufacturing. Whether deployed on camera or server, analytics can help factories tremendously. This note discusses how.
Video analytics can benefit various verticals, including manufacturing. Whether deployed on camera or server, analytics can help factories tremendously. This note discusses how.
Video surveillance is now part of the
Industrial Internet of Things trend. Cameras can help detect certain abnormalities, for example idling machines or workers not wearing safety gear. This then helps improve the security, safety and efficiency of the production facility.
A lot of this smartness has to do with
video analytics. These can be deployed on the server or
camera, which now has more processing power to support advanced analytics. “New smart camera capabilities have been brought about by the increasing amount of processing power available on cameras, which essentially turns them into IoT multi-sensors capable of analyzing data directly on the device,” said Anton Wintersteller, Senior Business Development Manager at Azena.
How video analytics can help
So how can video analytics help with factory operations? In essence they can play an important role in three areas. These are discussed as below.
Operations and management efficiency
Video analytics can increase operations and management efficiency, which in turn can help increase productivity, reduce downtime and ensure staff health at the manufacturing facility. “Smart cameras equipped with business intelligence applications can reduce downtime by foreseeing potential interruptions or production jams and alerting staff in real-time, allowing them to take action immediately. These same cameras can be equipped with apps that analyze overall production flow to evaluate specific bottlenecks or delays on the manufacturing floor,” Wintersteller said. “Facility staff can also use this information to optimize a manufacturing facility’s layout by analyzing potential obstructions or obstacles that would prohibit efficient production.”
“Analytics can be used to identify all employees wearing a smock, an image analysis that could be helpful distinguish between assembly line workers and those associated with another department,” said Jammy DeSousa, Senior Product Manager for Security Products at Johnson Controls.
Inventory management
At a time of rising export orders and increased productivity, manufacturers are placing more focus on
inventory management which can also be enhanced by analytics. “More goods than ever are being produced, packaged and shipped each day, and the higher the output, the more complex it becomes to monitor and quality-control. Smart cameras equipped with video analytic applications are used to manage large inventories in warehouse facilities, and in instances where cameras are not permanently installed in buildings, they can be located on drones that move autonomously through inventory and capture data by scanning barcodes,” Wintersteller said. “When goods and boxes of different sizes have to be stored with maximum efficiency in mind, analytics can help recognize incoming and outgoing items in real-time, according to size and format – enabling picking machines to sort them appropriately. Ultimately this optimizes storage and efficiency of space.”
Defects detection
This has to do with
machine vision, whereby deep learning-based analytics engines trained with thousands of images can help identify defects. This is a drastic improvement from human eye-based inspection and can help with manufacturers improve yield rates and quality control. “Much like a person’s brain that uses neurons to learn, artificial neuron networks are being used for image analysis in a production setting. For example, neuronet analytics can be used to identify that a box on a production line has been slightly damaged or has not been properly packaged,” DeSousa said.
Edge vs. server
There are several advantages of putting the analytics on the camera, which, as mentioned, now has more and more processing power. “Due to the recent increase in processing power on the edge, cameras are able to analyze video footage directly on the device, without transferring it to a video management system. This eliminates storing of irrelevant footage and increases privacy as only pertinent video will be transferred of the device,” Wintersteller said. “For example, if a smart camera is monitoring a manufacturing facility for potential break-ins and captures an unauthorized access by an individual after business hours, that event will be sent to security personnel for further action. All other video monitoring of employees and visitors entering and exiting throughout the day will not be stored. Not only does this increase privacy, but it requires less bandwidth as well.”
Yet there are instances where putting analytics on the server works better. The key then lies in the user’s own environment and use case.
“There are applications where video processed on the edge is the best approach and then there are other applications where video processed in servers works best. A good rule of thumb is that the algorithm is going to be more robust when it is run on a server than when it is run on a camera. For example, if your video requires more demanding applications that need to be accurate and efficient, then a server based application would be the best approach. However, if you are looking for scalability then an edge based approach would work best,” DeSousa said, adding there’s also the hybrid approach.
“There is a third option and that is a hybrid approach, where you can leverage robust algorithms for a few cameras out of a network of 30, and then rely upon video being processed on the edge for the remaining cameras,” he said.