top of page
PORTFOLIO
Thermal Camera Object detection and temperature calculation
Small TitleA Proof of concept has been build using AI in order to detect the person's temperature at real-time on Edge Device i.e raspberry pi(we have used raspberry, can be deployed on other as well). First, we have detected the object in real-time using SSD mobile net Caffe model with the backend of OpenCV and dnn which trained model on Caffe. Using Multiprocessing tasks, we have developed a model to make it run on Rpi. Then once the person has been detected, we took that ROI and find out the EYE region and calculated the pixel internity values at that point and taking a 25 degree of reference temperature and using a formula which I couldn't tell we have calculated the person's temperature and using flask & Django ad an API we have deployed the model on web
DeepEye a camera security product
Deep Eye is a camera security product made by me and my team This is the product that takes care of the security. This product consists of 4 different deep learning models such as face recognization, fight, no fight and fall detection, fire detection, and weapon detection. This product is the end to end product in which we have used Django along with the javascript, Ajax and cloud platforms in deep learning we have used Caffe, TensorFlow, Keras, Openpose,inceptionv3,
Deepsort ...etc.In this, the deep learning models have been integrated with the web. This product can make the security at the next level we have also integrated the alert system which sends the alert to a respected person. This product can be accessed by various levels
Superadmine level, Admin level, and Viewer level.
In this, the SuperAdmin Can add the multiple cameras in this system by just adding the Ip or the rstp link of that camera. The SuperAdmin can also make Admin and users and can assign them the camera similarly the Admin cannot add the camera but can assign the camera to the Viewer. At Viewer level, if he accesses the camera he will always have the access to the camera given by the Admin or SuperAdmin
Android Application for fire alarming system
In this I and my team have integrated Android with deep learning In this we have made a fire alarming system as soon as camera will detect the fire it will send the alarming message to the user who is using the application It will also tell the user about the minimum distance from the place where the fire happened . In this, for the fire detection, we have used Keras and Inceptionv3 and we have integrated the OpenCV code with firebase on other side our Android application is also connected with the firebase. In this application, we have provided with the user registration and many other functionality
CRM system using Django
I have made a simple CRM system on Django In this there are two app one for the customer and another for the vendor who can see the product and the customer details
In this, I have used SQLite database in which many to many and many to one relationship of tables is been maintained
I have also created the customer Login and signUp page
I this project I have used the concept of dynamic Urls for the smooth update and delete operations
The concept of connecting the tables in the SQLite database through the intermediate table
And I have also given the functionality to the customer about updating the order details and If the person forget the password then the customer can update the Password.
HRV and Respiratory rate analysis using webcam
Jump towards the Medical AI. A POC is supposed to build to make a cost-effective model for customers to analyze heart rate and respiratory rate using our web camera and mobile camera. I cannot disclose the exact details of this project but we are also developing the Android Application for the same. Heart Rate (HR) is one of the most important Physiological parameter and a vital indicator of people‘s physiological state A non-contact based system to measure Heart Rate: a real-time application using camera Principal is to extract heart rate information from facial skin color variation caused by blood circulation
Drone AI
A Proof of concept has been made in order to integrate embedded programming with AI or say IoT with AI. This is a project in which we have to integrate our AI or CV model to the drone using raspberry pi. The drone controller which we have used was ardupilot and we have connected it with raspberry pi using mavlink. The main idea behind this project is to make a security system for roads, traffics, prohibited are and so on.
ALPR detection using YOLO
A high jump towards a smart city. Our client needs to remove manpower and make the machine do
everything. He is the owner of a mall in Singapore and wants to develop a system that automatically
stores, car pictures along with the letters of the number plate from 15 cameras connected with the AWS
server. This system is used to detect license plates of cars exceeding a certain speed. We have trained an
object detection Model to detect license plate in real-time. This system is used for the detection of stolen and searched vehicles. The detected plates are compared to those of the reported vehicles. Once the plate has been detected, we will apply some basic thresholding operations to recognize the letters inside the plate.
A POC has been build to run a model on AWS windows server for real-time license plate recognition. The model which we have used is YOLO. It provides great accuracy and at last, we have to deploy it on the server so there wasn't a problem of computation
Student attendance system Using Computer Vision , Deep learning and android
we have made a student attendance system using Opencv,dlib, TensorFlow, and Keras
In this project, we have made a desktop application with the help of Tkinter
The UI made by Tkinter helps easily to take input from the students
First, we have created the dataset of the student which take the roll number the name, class and section on the basis of this the folder is been created corresponding to each student
Once we have made the date so we start training as there is only one picture for each student the model does not take much time to train
After training, we start the webcam of the computer/laptop or we can attach our mobile phone camera by using the Ip webcam through Ip address
Once we have the detection the student information with name, class, roll number and the time of entering the class to the firebase
From there we send the information to the mobile application where we have made the registration for the teacher so that they will get the information about the student
we also provide the update option for updating any information for the student through teacher
The mobile application on android and the desktop application are attached with each other through firebase
Centroid face tracking system
I have developed a model with Caffe as the platform and SSD as the model and the mobilenet as the architecture for the project In this I have trained the 700 face images having more than 1500 instance we have trained this model After than I have deployed this on raspberry pi 4 this model is giving me 10 fps on the rpi-4 after making the processing parallel It give me 15 FPS
For increasing the speed to 25-30 FPS I have used Intel computational stick
bottom of page