IoT and AI can be consumed in the realm of Digital Transformation through Computer Vision as a technology enabler.
For Example: In Buildings today, people install CCTV cameras and devices to track and monitor the people flow within and around a building premise. While most cameras today only capture data, however with the inception of Computer Vision, we can have smart and intelligent camera devices that can scan, detect and identify objects in realtime using Computer Vision technology, synthesise/process the captured information on Edge capable devices for both online and offline data share and send pre-emptive alerts on suspicious people flow.
Today most buildings are thriving for Smart Building ecosystem and infrastructure readiness, and this sort of an amalgamation of Computer Vision coupled to IoT/Edge based smart devices, gives real time edge based enhanced security, which can be absolutely crucial and critical for the residents.
To enable such smart monitoring ecosystem, the first step would be to have the smart cameras enabled which typically come with embedded Edge based cameras to capture live stream of images/videos. Once an image gets captured it ideally needs to start a series of processes for detecting objects within the image and then further classify the image verifying their identities.
Object Detection is a complex process that involves Deep Learning customized consoles and libraries and today most commonly used platform happens to be Tensorflow/Pytorch. Tensorflow easily integrates with cloud infrastructures and also supports Serverless topology (such as AWS Lambda). To simulate the data models and have them synced with IoT stream, integration API’s are usually developed on Python.
Once we are able to set up the data model using Tensorflow Libraries and have the engine integrated with the Smart Devices, we will now have to train the data models to ensure trusted and fair output results, from the continuous data influx stream from the Edge camera devices. Training the Data Models is a continuous process that involves steps such as Verifying the API installation libraries, Downloading fresh stream of Live Images, creating xml files for each image, converting the xml’s into csv’s, run the python files, test against the data modeling libraries from TensorFlow, map the output against graphical interface and continue testing.
Typical Output looks like the following:
Once this is accomplished and we are able to detect various objects within an image file, the next step would then be to verify identities of the people among the objects detected. This is where we apply Image Classification metrics.
A combination of IoT/Edge mapped with AI layered with Object Detection and Image Classification can create a highly sophisticated Smart People Flow Monitoring system. Same technology flow can be applied to every possible industry that one can think of, with aspects like Predictive Maintenance, Offset Detection, Anomalies Detection, Critical Machine Failures, etc.