Restricting the scope of video analytics to merely surveillance and security-based applications would do gross injustice to the development of this field of engineering. Video Analytics should not be viewed as a solution that is limited in application due to the boundaries created by physical devices (like CCTV cameras) in use.
The central agenda should be to utilise the video analytics to gather “good” data and perform an operation based on analytics of that data. Eventually, the video feed should be engineered to provide data with the same credibility as a physical sensor deployed in critical areas does. These solutions can find a lot of applications in industrial operations optimisation.
One area which will provide a lot of scope to video analytics software to offer solutions in the area of operations. Generally speaking, if an operation is executed on a specific repetitive pattern, there is a scope to automate it. The overall architecture of such a system is quite similar to a conventional feedback system where the process variable drives the system. In this case, the data set obtained from the video feed drives the response system.
Data acquisition: The stepping stone in designing a good system is the ability to differentiate and gather “good” data. This is the most important step in the design process since all analytics algorithm outputs are only as good as the data they operate upon. In the industrial context, there are two types of operation optimisations that can be taken into account.
First is the overall optimisation technique based on work study, time study and motion study, which finds more relevance in heavy industries involving heavy equipment and machinery and movement of personnel around them. Second is a micro level analysis of physical movements and gestures of people involved in specific operations.
As far as data gathering is concerned, it should be taken into careful consideration to acquire data from processes that are repetitive in nature. Here the physical placement of cameras also plays an important role.
Transforming data into operable entities: The next step in the design is the conversion of the data acquired into values and entities that can be used to drive an engineering system. This is done by assigning attributes to the data acquired. As an example, consider the case of personnel movement on the shop floor.
The video feed can detect motion in the area of interest and provide specific contours of the personnel involved. However, this data is not enough to understand patterns of movement on the shop floor. An extra attribute has to be added to these contours to determine the nature of the movement. Unless this attribute is added, the data acquired will not be helpful in designing an intelligent system.
Data segmentation: Once the data of the repetitive processes has been acquired, it is time to put the data into bins which are homogenous in nature. The purpose behind this activity is to aid in the definition of the ideal way of performing a certain repetitive activity. In the absence of this segmentation, the acquired data will represent a behaviour too random to be logically discernible.
Pattern establishment: Using the attributes assigned to the data, it is possible to assign tags to data sets. These tags are primarily responsible for establishing patterns from the data sets. As far as the operation optimisation is concerned it is the analysis of these patterns that enable the software to understand the behaviour of operations. Depending on the requirement, data analytics has to be used to appropriately to make the response of the system as accurate as possible.
Definition of ideal behaviour and violations: Once the data segmentation and organisation has been done on the basis of attributes assigned to the data, it is possible to assign tags to the patterns associated with the data sets. For example: Consider a traffic management system which identifies the instances when cars are not following lane discipline.
Here the data sets are comprised of locus traced by individual vehicles. These traces will, in general, be similar in character. Using the mathematical tools a differentiation between these patterns can be established. After this differentiation is achieved the next step is to assign tags to these patterns so that ideal behaviour and violations can be accurately determined.
Designing response system: The response system shall be a pretty straightforward design as far as the trigger system is concerned. The complications lie in the usage of data analytics involved. To understand this better let us understand what analytics means to the response system. In the case of large manufacturing industries, where physical movement of people and assets is something that can be improved for operations optimisation, the dataset derived from video analytics should be able to provide a detailed description of the patterns of movements that emerge under regular and irregular circumstances.
These patterns should then be mapped to other dependent variables in the production unit e.g maintenance of equipment. Based on other data mapping techniques, a correlation should be established between the patterns of movement and productivity indicators e.g. daily output. In order to optimise any operation, it becomes important to understand that operation in complete detail, taking all aspects into consideration.
To be able to aid in industrial operations optimisation there are two primary solutions that can be developed once the data processing is complete. These are:
On the basis of real-time feed: Gross violation and deviations can be notified immediately and corrective action can be designed into the system. The violations are determined in real time on the basis of the data segmentation and pattern establishment that is done when the response system is designed. In principle, this system is quite straightforward and is similar to a basic event-triggered system.
On the basis of historical data: This is more comprehensive and involves some aspects of artificial intelligence to be brought into the picture. Based on the comparative analysis of data sets obtained from conventional patterns of operations and the data sets (created using predictive algorithms or obtained under test conditions) representing optimised conditions, optimisation solutions can be designed. This system should ideally have an advisory role as even the best systems might miss out on certain environmental conditions or unexpected events which might make the behaviour of the system unpredictable.
In consonance with Industry 4.0, industrial operations optimisation is an area where there is immense scope for growth and improvement.
To be able to successfully design optimisation solutions for the industry an interdisciplinary approach involving video analytics, image processing and data analytics has to apply. A detailed description fo each intermediate step mentioned will be presented in the blogs to follow.