/indoor-wifi-positioning

Primary LanguageJava

Indoor Wi-Fi positioning

We suggest Wi-Fi positioning based on Fingerprinting

  • Figerprinting - A unique characteristic of the radio signal (e.g. Identify a position based on a set of RSSIs associated with APs

Goal

Develop Android application for indoor positioning (AI 공학관 4, 5층) using Wi-Fi and sensors

Blueprint of AI 공학관

스크린샷 2022-05-09 오후 9 53 12

1st week

Progress

  • Set the role for each member
  • Create Admin App
  • Implement Wi-Fi sensing
  • Create User App

To Do

  • Collect data using another sensors (e.g. GPS, Barometer)
  • Implement back-end server using firebase realtime database
  • Implement python environment for analysis sensor data

2nd week

Progress

  • Make an Admin app that can measure the RSSI value of Wi-Fi
  • After uploading the measurement value to the database (FIREBASE), convert it using Python

Admin measurement app testing

Use the spinner to select the number of floors and and press the START button to find the RSSI(Received Signal Strength Indication Value) value of nearby Wi-Fi and inform the WAP(Wireless Access Point) and distance.

Wi-Fi Signal Stength

Contains the RSSI of the WAP at the measured location. When RSSI values are expressed in negative format, the closer the value is to 0, the stronger the received signal.

WF1

Data exploration

image

Only the reception strength of the measured value is measured in the (FIREBASE) data set. Convert the measured value to a data frame using Python

#원하는 호 수 => 504호 => (5층/4호)
ref = db.reference('?층/??호')
p = ref.get()
get_list = []

df = pd.DataFrame(p)
df.index = np.arange(len(df))

#호 수로 입력
df["target"] = ???

df

image

3rd week

Progress

  • System Architecture
  • Model evaluation

System Architecture

  • Location data collection image If possible to measure the interior of the classroom, measure approximately 30 randomly.
    If impossible to measure the interior of the classroom, measure approximately 30 times at random by dividing the area between the rooms accurately.

Model evaluation

  • Decision Tree without preporcessing.

  • image Accuracy: 64% => Data need to be preprocessed!

  • Data preprocessing - Augmentation
    : Techniques used to increase the amount of data by adding slightly modified copies of already existing data.

  • image We covered each area with N(0, 3) for classroom, N(0, 2.3) for corridors according to the normal distribution and summed up with original dataset.

  • Data preprocessing - Data filtering
    : remove some unwanted components or features from a signal.
    Some observation is sparse.(For example, individually operated hotspot..)
    We regarded it as useless.
    We drop observations consisted of 45 percent missing value. image

  • Data preprocessing - Hyperarameter tuning
    RandomForest with GridSearch: remove some unwanted components or features from a signal. image

After data preprocessing

image

4th week

Progress

  • Dataset collection
  • Dataset Reorganization
  • Implement restAPI using google AI platform
  • Associate information with user apps

Release to User App

  • Using Google AI Platform:
    AI Platform is a service created by Google Cloud Platform that helps you implement and deploy machine learning models for production
    image
  • Tools to help implement, deploy
    • Serving, TFX, Explainable AI, TPU
    • AI Hub, Kubeflow
    • Various tools are available and are increasingly being added

RestAPI

image

After deep learning the dataset through the Google AI Platform and completing modeling, it is released using RestAPI
It uses Google's Identity Platform Rest API to perform common user tasks, such as user token tasks.

Check Location

image image

Deployment

Built With

  • minSdk:21
  • targetSdk:31

Authors

License

This project is licensed under the MIT License - see the LICENSE.md file for details

Acknowledgments

  • Knowledge about 802.11 protocol
  • Basic android programming skill with java
  • Basic Python programming skill
  • Data analysis technique