| room 213 | features |
BeepBeep-attendance is a project for accurately tracking the location of students seated in a classroom at the seat level, developed as the final project for the Convergence IoT Project course at Chung-Ang University. After reading several papers on indoor localization and conducting experiments, BeepBeep was selected for its superior performance and rigidity to environmental changes. In experiments conducted in room 213 of building 208, all 10 seats showed accuracy within a range of 20cm, achieving seat-level accuracy. This project earned the highest score among the teams during class presentations.
The project "BeepBeep-attendance" is based on the paper titled "BeepBeep: A high-accuracy acoustic-based system for ranging and localization using COTS devices" by Peng, C., Shen, G., and Zhang Y. in 2012. For specific details regarding the principles and methodologies employed, I recommend referring to the paper itself.
Refactored the project I’ve developed in the Fall 2023.
- 4 Android devices for transmitting/receiving chirp signals: 1 for the client and 3 for the beacons.
- Install the app on each Android device: This app receives commands from the command server and transmits/receives chirp signals accordingly.
Uses docker-compose and Makefile for fast build.
make local| Place the two speakers (microphones) facing each other | The client repeats 3 times for each beacon |
- Beacon Placement: Place three beacons evenly around the client. Since the client's location will be determined using the beacons, ensure they are evenly distributed.
- App Launch and Message Queue Connection: Connect all beacons and the client to the Message Queue server.
- Physical Alignment: Elevate one of the beacons and the client device to avoid obstruction and ensure that their speakers and microphones face each other.
- Command Server: The command server issues a chirp signal command to the facing beacon and client.
/commands/both - Sound Signal Transmission: The client emits a 1-second chirp signal, followed by the facing beacon emitting a 1-second chirp signal.
- Recording and Transmission: Both the emitting beacon and the client record for approximately 5 seconds and transmit these recordings (recorded at a sample rate of 48kHz) to the Message Queue server.
- Distance Measurement: The command server retrieves the recorded data from the MQ server and saves it in .json and .wav formats. Subsequently, using Matlab, it measures the distance between the two devices using methods like finding and cross-correlation.
/calculate/beepbeep - Iteration: Repeat the above measurement for all beacons.
- Triangulation: Using the positions of the client and three beacons, determine the client's position through triangulation.
/calculate/triangulate - Final Position Determination: Determine the client's final position by selecting the closest table based on the measured position values.
/calculate/locate
Finally, BeepBeep estimates the client seat like below.
Docker: Streamline project environment setup among team members. Given the diversity of individual development environments and the fact that some team members may lack experience with server setups, configuring the environment can be challenging. By usingdocker-compose, we've made it easy to set up experimental environments with just a few commands.FastAPI: Has Built-in interactive API documentation (Swagger), eliminating the need for separate frontend development to issue commands to the command server. Through Swagger, commands can be interactively sent to the command server, simplifying the process without the need for additional frontend work.ActiveMQ: Advantageous in scenarios where there is a need for bidirectional communication between one command server and multiple beacons, with varying beacon counts or the addition of broadcast functionalities like 'stop all' commands. In such cases, introducing a message queue is beneficial for managing communication efficiently. (Another message queue also can be used)MATLAB: although Python libraries like SciPy and NumPy can be used for BeepBeep(cross-correlation needed), we opted for MATLAB because it was more than 3 seconds faster.
- GwanBin Park: Backend (all of these!) and Testing
- Joohee Cho: Paper analysis and Testing
- Taekwan Nam: Android (sensor data collection) and Testing
- UiChan Jeong: Paper analysis and Testing
During the demonstration, we ran a total of 10 tests. When the data came in successfully, each measurement provided accurate distance estimates (within a maximum error of about 20 centimeters based on Euclidean distance, with an average of about 10 centimeters), and the measurements were also successful in edge cases, such as corner spots and spots farthest from the beacon. This project earned the highest score among other teams during class presentations.
- Increased error due to obstacles: Accuracy would significantly degrade due to reflections if obstacles were present between the client and the beacon. If Android speakers are not facing each other, inaccurate distance measurements may occur due to sound reflection. However, aligning the direction of speakers may be difficult in practical environments.
- Decrease in accuracy due to noise: When ambient noise interferes with the signal, the accuracy of location determination decreases. Test results showed that simple ambient sounds didn't introduce significant errors, but loud noise or noise similar to chirp signals degraded accuracy.
- Need for manual command execution: Although attempts were made to implement a program to automatically rotate devices, automation couldn't be achieved due to limitations in the number of usable devices. However, it's expected that automation would be feasible with sufficient device availability in the future.
- Discomfort caused by audible frequencies: While using audible frequencies yielded acceptable recording quality for smartphones, generating chirp signals in the audible frequency range may lead to discomfort during actual usage.
- WiFi RSSI: Initially, it was expected that WiFi-based RSSI measurements would yield high accuracy. However, in reality, even when measured from the same location, the values varied within a range of 1 to 15 dB, making it impractical for actual use.
- EchoTag: Using EchoTag: Yu-Chih Tung and Kang G. Shin, Accurate Infrastructure-Free Indoor Location Tagging with Smartphones was also considered. Tests were conducted to determine if table positioning could be identified using the EchoTag method at each table. The results showed that distinguishing between tables blocked by walls and those between the middle table was somewhat feasible. However, distinguishing between the front and back tables in the center was nearly impossible. To address this issue, the possibility of placing objects on both sides to facilitate artificial reflection was considered. However, consistently placing identical objects in the exact same positions within the classroom was deemed impractical, leading to the exploration of alternative methods.
- Computer Vision techniques are not allowed in this project.