InsightAI is an artificial intelligence powered software which helps the retailers to grab useful insights and monitor them in real-time so as to enhance customers' shopping experience. During this ongoing pandemic crisis, the retailers need to be more concerned about the maximum limit of people in the store or a specific area and also maintain social distancing between them to follow the safety norms. The retailers can monitor this in real-time as InsightAI has built-in automated algorithms which monitors the people count inside the store and also checks social distancing between people inside the store. The retailers can also specify the maximum threshold for this, which on getting exceeded would automatically initiate an alert email to them in real-time. Apart from this, InsightAI has various other features like, an excel sheet gets created at the end of each session which can be used for footfall analysis, automatic enabling of timer and session scheduler etc. InsightAI is built so as to accept diverse set of arguements during runtime. It is a scalable software and is capable of running over varied inputs like pre-recorded videos, live video through a webcam or even a connected ip camera. InsightAI has a built in video writer which can be used by the retailers to permanently store output videos to local disk.

---

Check out InsightAI's product landing page here: https://insightai-10.web.app/

---

InsightAI software modules :

• Store Entry/Exit Analysis
• Instore Crowd Analysis

---

(1) Store Entry/Exit Analysis :

InsightAI monitors and analyzes the store's entrance/exit point. It counts the number of people entering and leaving the building and the total number of people present in the building at any instance and provides realtime analysis in the form of a GUI. It also sends an email alert if the number of people present in the building at that moment exceeds a given threshold. An excel sheet also gets created at the end of the session which contains some quite useful data that can be used by the administrator for footfall analysis. The administrator can also setup the timer (session's max time limit) and scheduler (automatically starts the session at the scheduled date and time). Output video gets generated and stored permanently on local disk and can be viewed by the admin anytime.

Environment used:

• Anaconda
• Jupyter

Packages and dependencies:

• numpy
• pandas
• scipy
• imutils
• datetime
• schedule
• opencv-python
• dlib (You need to install cmake globally prior to dlib installation)
• jupyter notebook
• ipykernel

Implementation steps:

• Virtual environment setup in anaconda
• Obtaining input video (prerecorded video or webcam)
• Image preprocessing and blob inferencing
• Training the MobileNetSSD neural network
• Construction of centroid tracking algorithm
• Creating a trackable object
• Implementation of people counter
• Writing output frames to a video file on disk
• Implementing people's log file, mail alert, timer and scheduler setup and max threshold functionalities

Features:

• InsightAI is capable of running over both input videos as well as in realtime through webcam or ip-cameras, giving a higher FPS throughput rate than even yolo and r-cnns
• The software user (or) the store administrator can even enable mailer functionality. Doing this, they will receive an email alert if the number of people present in the store exceeds the specified threshold.
• I have also provided a functionality, which on enabling results in the automatic creation of a log file (excel sheet/csv file) at the end of the session, which would be helpful for footfall analysis.
• The user can also setup the timer upto which the video runs and then stops automatically.
• Keypresses are also captured and are handled correspondingly.
• Video summarization text is displayed on the window.
• Output frames can be written to a video file on local disk for permanent storage(if the user specifies --output arguement during run-time). The output video can be recorded in any supportive format like avi, mkv, mp4v.
• Automatic cleanup work is also done at the end of the session.

Output of Entry/Exit Analysis module:

• Basic run-time anaconda terminal:

Specify the --input arguement leading to the corresponding path of the prerecorded input video (or) if you want to enable webcam specify url=0 in the config file. Note: Arguements can be manipulated in Main.py as per your convinience

• Config file:

Note: The config values can be manipulated manually as per your requirements.

• Mailer functionality:

This functionality can be enabled by specifying the mail id on which you wan't to receive the mail in the config file, setting up the maximum threshold and specifying ALERT=True

As soon as the maximum threshold you specified gets exceeded, you receive the corresponding mail.

• Log file functionality:

Set Log=True in the config file

• Timer functionality:

Set Timer=True in config file and change the time limit as per your requirement in Main.py as shown below,

• Scheduler functionality:

Set Schedule=True in the config file and then specify the schedule time and date in Main.py as shown below,

• Realtime analysis in GUI:

The above image is just a screenshot of the video being displayed as a GUI in realtime.

---

(2) Instore Crowd Analysis:

InsightAI monitors the crowd inside the store. The admin can specify the ROI limit which indicates the area needed to be monitored for analysis. It detects the movement of people inside those the store, calculates the distances between them, analyzes the people violating social distancing and provides the reults in the form of a GUI to the adminstrator in realtime. The admin also gets an email alert if the total number of people violating social distancing exceeds a given threshold. As in Store Entry/Exit analysis module the admin can set the timer & scheduler and can also access the output video anytime later from the local storage.

Environment used:

• Anaconda
• Jupyter

Packages and dependencies:

• numpy
• pandas
• scipy
• datetime
• schedule
• opencv-python
• jupyter notebook
• ipykernel

Implementation steps:

• Virtual environment setup in anaconda
• Obtaining input video (prerecorded video or webcam)
• Frame differencing and image preprocessing
• Training the YOLO neural network (Note: You need to download YOLO weight file and paste it into the model folder inside Instore_Crowd_Analysis. Its file size is around 250MB and hence I couldn't upload it in the repository. You can dowload the weight file from the given or any other preferred link. https://pjreddie.com/media/files/yolov3.weights_ )
• Construction of centroid tracking algorithm
• Creating a trackable object
• Implementation of social distance detection algorithm
• Writing output frames to a video file on disk
• Implementing mail alert, timer and scheduler setup and max threshold functionalities

Output of Crowd Analysis module:

• Basic run-time anaconda terminal:

Note: Arguements can be manipulated in Main.py file inside Instore_Crowd_Analysis folder as per your convinience

• Config file:

Note: The config values can be manipulated manually as per your requirements.

• Mailer functionality:

This functionality can be enabled by specifying the mail id on which you want to receive the mail in the config file, setting up the maximum threshold and specifying ALERT=True

• GPU available:

You can enable the GPU in the config file. This will speed up the frames' analysis and subsequently reduces the processing time. (This can only be done if you have CUDA NVIDIA GPU)

• Time and Scheduler:

Set Timer=True and/or Schedule=True in the config file.

• Realtime analysis in GUI:

The above image is just a screenshot of the video being displayed as a GUI in realtime.

---

System Architecture of InsightAI:

---

User Manual to run InsightAI on your system:

• Installing the tools:

  • Anaconda Navigator:

  Install anaconda navigator from its official website.

  • Python:

  Install python 3.6 version from its official website.

• Setting up a virtual environment:

  • Create an environment in your anaconda prompt using the command;

    $ conda create -n env_name python=3.6

  • Activate the environment with;

    $ activate env_name

  • Install all the necessary packages and dependencies:

    $ pip install package_name

• Download the software files from my repository

• Launch The Jupyter notebook

  Launch the Jupyter notebook from your project’s working directory. For this navigate to your folder's location from the anaconda prompt by:

  $ cd path_name

  $ jupyter notebook

• Setting the Mailer functionality:

  Specify the email through which you want to send the alert mail in Mailer.py file. Make sure that the specified mail id is a 2 step verified account. Now, you need to generate a 16 character app password and paste it in Mailer.py file. To generate the app password, refer to the guidelines present in the following link;

  https://support.google.com/accounts/answer/185833

• The Config File:

  The config.py file can be altered by the admin as per his requirements. This file can be found in the mylib folder. The admin can easily specify the values for url (0 by default for webcam, or specify the url of the ip camera connected to it, or pass a pre-recorded video during runtime), mail id, alert, threshold, scheduler, log, and timer.

• Running InsightAI:

  After successfully setting up InsightAI, it can be executed from the anaconda prompt(you need to be present in the same environment and path);

  • Store Exit/Entry Analysis:

    $ cd Store_Entry_Exit_Analysis

    $ python Main.py --prototxt mobilenet_ssd/MobileNetSSD_deploy.prototxt --model mobilenet_ssd/MobileNetSSD_deploy.caffemodel --input videos/input/example1.mp4 --output videos/output/ovideo.avi

  • In-store Crowd Analysis:

    $ cd Instore_Crowd_Analysis

    $ python Main.py --input videos/input/crowd.mp4 --output videos/output/ouvid.avi --display 1

---