AI Algorithms, Software, Cloud and Edge Processing Analysis

Purpose of the Analysis : Public perception of the AI, software, cloud and edge processing concepts

Purpose of the Analysis :

Public perception of the AI, software, cloud and edge processing concepts tend to be broadly confused and misinterpreted by everyday users. In fact most of these technologies are marketed to the users within the service providers’ product or service capabilities. This brief analysis emphasizes on the fact that the referred terminology has a greater meaning and usage.

This is a simplified and limited scope post. We are able to provide further stats, explanation and custom solutions. Please contact us to discuss your business case further.

Software Engineering

Software engineering is a design approach to architect a system using programming languages to perform computing capabilities (data collection, data storage, data processing, data presentation…etc.) on various resources. Software can be developed to run on variety of environments and resources as mentioned below.

IMPORTANT : Resources are computing capabilities, virtual machines, physical computers including but not limited to the hardware in your office, cloud provider, UAVs, UGVs, your car, satellites, remote stations, nano computers, mobile devices, TV, embedded in a fridge..etc. whatever you can imagine…

Expand your imagination beyond the capabilities of the widely marketed public cloud providers.

Cloud Computing

Cloud computing does not mean moving your servers to a public cloud infrastructure providers such as AWS, Azure…etc. These providers are tremendous benefits to users ; 1) Simplicity / Ease of use, 2) On-demand availability variety of computing resources, 3) operational capabilities…etc and more, however many individuals and businesses think within these providers’ capabilities and scopes.

Cloud computing is a designed architecture of your internal and external software and hardware resources in a hybrid scheme to maximize the resource utilization (data sources, assets, connectivity, data processing and visualization capabilities).

Edge / Asset Computing

Cloud computing does have many barriers and challenges that are not disclosed or outspoken often. Using a central data structure, even though distributed to multiple regions, may impose high bottlenecks to users or expected results. Especially when the field data sources such as UAVs, UGVs, vessels…etc. are out in a low data connectivity areas(mountains, ocean…etc.), then even if the asset can collect TBs of data, the bandwidth may not be enough to send all the raw data. In such a scenario, Connected Ops Edge architecture, including APIs and internal commands, provide processing capabilities on the asset itself. These capabilities are combination of imagery sampling, data classification, temporal and spatial analysis and more. There is no single perspective or approach to a specific challenge, therefore Cloud Ararat can help you analyze the trade-offs for a realistic solution.

Artificial Intelligence

Artificial intelligence, machine learning and deep learning terminologies have been popular among various industries as the computing capabilities increased exponentially and widely available to public at affordable prices. There are open-source algorithms as well as proprietary technologies, where some of them are derivatives of the open-source ones.

IMPORTANT : The object detection approach used in this benchmark test, YOLO, was submitted by Joseph Redmon, Santosh Divvala, Ross Girshick, Ali Farhadi in 2015. We derive the algorithms from the referred YOLO algorithm, however the instead of pre-trained data sets, we can develop custom implementations for your business.

Our capabilities are not limited to presented content in this post. This is only a limited representation of demonstrated hands-on capabilities.

Key Statistics and Observations

Most Variety of Classification : Among the tested algorithms, YOLO 9000 data set has over 9000 different classifications available. The algorithm, being the slowest, processed 120MB of static video file in approximately 32 minutes. 11GB GPU with 56GB ram and 500GB SSD computing power has been utilized for static data processing. Live streaming performs at 3-7 frame per second rate without a lag over a regular 4G connection. GPU is a must when performing image processing capabilities. Virtual machines tend to perform slower than dedicated physical device.

Faster Result : Yolo V3 Large data set had fast performance and relatively great results. It took 3 minutes and 30 seconds, compared to 32 minutes using Yolo 9000, for the same video and configuration. Relatively for generic scene such as city view, data set with 80 trained objects provide realistic and high confident(over 75%) results. Especially smaller and distant objects were identified promptly. Over a live streaming, algorithm process at a 25+ frame per second.

Balanced Results : Open Images took 10 minutes to process the same data under same conditions compared to 3:30 minutes for V3 large and 32:30 minutes to 9000. Open Images have 601 classified objects, where the detection triggers a sub-category after a given confidence threshold the programmer defines. For example first identify a car in a bounding box, then sub-cat vehicle, followed by a land vehicle as the 3rd inner layer of bounding box. We had to tweak the source code of the algorithm to be able to maximize the computing resources(core usage, ram usage).

Manage Computing Resources using Connected Ops Command & Control Distribution Center: Depending on the business case requirements, following top 3 priorities; 1) data sources, 2) data connectivity to the sources 3) computing capabilities at the data source will have the highest impact to the design factors of the overall architecture. If the data connectivity is low we prefer to utilize edge computing on the data resource(satellite, UAV, UGV…etc.) 1) process the sampled data, 2) generate actionable intelligence output data, 3) transmit only the actionable intelligence data structure to another asset directly(asset-2-asset) or to the central repository.

Connected Ops is an Innovation Platform

Software as a Service : We provide entire solution with web, mobile and asset applications for your business ecosystem. We can create a custom solution for your business using Connected Ops, which would drastically reduce “Go-live" period. Do you need a solution? Contact us to discuss. Our vision is to make high-tech solutions available to non-tech people

Integrated Platform as a Service: Extensive API availability for interacting with the internal and external resources. Platform can be integrated to your preferred assets (UGVs, UAVs, manned systems, satellites…etc.), business systems, accounting systems, CRM and more. Please get in touch to design a solution together.

Infrastructure as a Service : We are not a re-seller of any public cloud provider, therefore we deploy the Connected Ops solution to an environment that is right for your business. AWS, Microsoft Azure or your own data center. Data systems can be in a single cloud environment as well as distributed to a multiple cloud environments. We ensure that the relevant business continuity and disaster recovery scenarios are implemented for your business case.

Would you like to cooperate? or hear more about our solutions?