BSafe

Cloud Flow Prediction and Client Discovery Using Wireless Networks

Introduction:

Due to population growth, crowd analysis has become a major interest in social and technical disciplines.Crowd analysis is being used to develop crowd management strategies in public events as well as public space design, visual surveillance and virtual environments to make areas more convenient in order to prevent crowd induced disasters.

Identifying crowd patterns in a sample setting (IIIT-H) by WiFi requests sent by mobile devices as they move around. The client locations can be triangulated to know their accurate locations at any point of time. This data will be used to create heatmaps and perform time series analysis

Every device, which has Wifi ‘on’, performs ‘Active Scanning’ where it continuously transmits probe requests, which consists of BSSID (Broadcast MAC address), SSID (Zero Length) and MAC (device MAC address).As MAC Addresses are globally unique, hence tracking the movement of a particular device is possible.With a MAC address, these logs can be cross referenced to track the device in all of the places which has a listener. If a listener is driven around a residential area, then one can know the location of the house from which the device is from.

How Wifi Works?

1. Turn the receiver on and listen on each channel for the beacon.
2. Broadcast a "Who is there?" packet on each channel.

In (1) battery life is negatively impacted and 

power usage for a wifi adapter sky-rockets.

So we use (2) In this process, the wifi card 

transmits a probe request. The probe request consists of:

BSSID: Broadcast MAC address, SSID: Zero length, MAC: Your wifi MAC address

How we exploit this?

• We set up a device to purely sit silent and listen for these probe requests, then write the MAC address and timestamp to a file. 
• So we can tell when you pass within range of my access point , for as long as you are in the area. 

Underlying Math Involved in Triangulation:

Analysing the IIIT setting

Building a Heatmap of Crowdflow

a)Heat distribution over IIIT campus b) Corresponding satellite image

• Setting up Raspberry Pis at multiple location thorugh out the campus.
• Collecting and pre-processing the data.
• Using the data collected by the Raspberry Pis, we will build a heatmap of crowdflow.
• The red circles shows the hotspots of crowd averaged over all time in IIITH over a time period.

Time series analysis of Crowd Flow

• Time series analysis is the process of forecasting time series data based on an additive model where non-linear trends are fit with yearly, weekly, and daily seasonality, plus holiday effects.
• We utilize two python libraries to do this - Prophet and Arima (AutoRegressive Integrated Moving Average)

• The four graphs show performance of Arima w.r.t to ground truth:
• Arima Predictions in 2-D based on Training data
• Ground Truth of Actual Path taken by Target in 2-D
• Arima Predictions in 3-D based on Training data
• Ground Truth of Actual Path taken by Target in 3-D

Web Interface:

Live web interface collecting and parsing data in real time and generating predicted crowd flows and analysis graphs in real time.

 

Anomalies and Abnormal Crowd Behaviour:

• When there was any important event on campus, we saw that there was an appropriate response in the crowd movement. 
• We can see  in the above images that when there was the "R&D showcase", the crowd was concentrated at KCIS and nearby regions. 
• During the "Farewell" that happened a few days ago, there were a lot of people at the Felicity Ground and nearby regions.

Extensions to this Project:

• We can analyse bigger crowd settings like that of a big city like Hyderabad by deploying more Raspberry Pis at optimal locations throughout the city.
• We can extend this project by analyzing the unencrypted packets being shared over WiFi. There are architectures such as CreepyDol, one might get access to sensitive personal data, which can be used to monitor people's actions remotely. 

References:

1. ARIMA - Autoregressive integrated moving average.
2. Facebook, Prophet - Forecasting procedure implemented in R and Python.
3. https://linuxnet.ca/ieee/oui/,Make of the device from its MAC ID.
4. Chuan-Chin Pu,Indoor Location Tracking using Received Signal Strength Indicator.
5. https://www.raspberrypi.org,Raspberry Pi Setup.