Competitors

Team name: THWS & cronn

Team members: Toni Fetzer

Affiliation: Technical University of Applied Sciences Wuerzburg-Schweinfurt

Description:
The indoor localization system used in the competition employs a particle filter based on the CONDENSATION algorithm. The system utilizes Bluetooth low energy beacons, Wi-Fi, and an inertial measurement unit (IMU) in a commercial smartphone to gather data. The process includes three main stages: transition, evaluation, and resampling. The transition stage updates particle positions using IMU data, while the evaluation stage assesses these positions based on signal strength models for Wi-Fi and Bluetooth. Resampling addresses weight degeneracy by selecting particles based on their weights. The system ensures real-time performance and can handle sensor outages.

Team name: ATR

Team members: Qi Zhang, Lingxi Zheng, Shanshan Zhang, Lingxiang Zheng, Ao Peng

Affiliation: Xiamen University

Description:
We have adopted an Inertial Network aided Visual-Inertial Odometry. In this system, VIO serves as the core positioning model, while the neural network-based IMU distance estimation and geomagnetic data provide comple mentary inputs. These inputs are fused using factor graph optimization to improve overall accuracy and robustness in challenging environments, such as low-light or low-texture areas. This fusion approach ensures a more reliable and consistent indoor navigation experience.

Team name: SZU-Escope

Team members: Mengyuan Tang

Affiliation: Shenzhen University, E-scope

Description:
Our system addresses the challenge of indoor navigation in multi-floor buildings where GNSS signals are weak or absent. Using Pedestrian Dead Reckoning (PDR) with smartphone sensors like accelerometers, gyroscopes, and WiFi signals, it estimates pedestrian positions and recognizes behaviors like stair climbing and elevator use. The system combines Kalman and Particle Filters for accurate trajectory estimation, and WiFi fingerprinting for location determination. Additionally, barometer and WiFi data are used to estimate the current floor level, enabling precise multi-floor navigation. This integrated approach provides robust, real-time positioning in complex indoor environments.

Team name: IA3

Team members: Alex Martínez-Martínez, Cristian Abundio, Raul Montoliu

Affiliation: Universitat Jaume I

Description:
Our indoor positioning system achieves high accuracy and reliability by integrating multiple technologies and methods. It uses multi-sensor fusion (accelerometer, gyroscope, magnetometer) and WiFi fingerprinting to predict user position with precision. Pedestrian Dead Reckoning (PDR) predicts movement based on displacement, while Long Short-Term Memory (LSTM) models, along with Recurrent Neural Networks (RNNs) and Convolutional Neural Networks (CNNs), are employed for advanced predictions. Activity recognition methods detect floor changes and turns, refining location estimates, and map information ensures navigation within building boundaries. This combination of techniques and models ensures the system delivers robust and accurate indoor positioning.

Team name: SZU-Escope

Team members: Mengyuan Tang

Affiliation: Shenzhen University, E-scope

Description:
Our system addresses indoor navigation challenges in multi-floor buildings where GNSS signals are weak. We use Pedestrian Dead Reckoning (PDR) with Inertial Measurement Unit (IMU) data, alongside deep recurrent neural networks, to process sensor data and BLE beacons. This approach enables accurate trajectory positioning by predicting pedestrian behaviors like stair climbing and elevator use. The system combines these predictions with Kalman Filter (KF) algorithms for precise, real-time trajectory estimates. This method offers a robust, low-cost solution for navigating complex indoor environments where traditional GNSS-based systems fail.

Team name: Indoor Navigation Team-CETC-54

Team members: Shiyuan Liu, Baoguo Yu, Lu Huang, Heng Zhang, Jingxue Bi

Affiliation: Navigation National Key Laboratory, CETC-54

Description:
Using WiFi and Bluetooth data to construct a 2D map, build a convolutional neural network model to establish mapping relationships, and integrate the positioning information derived from IMU data to obtain the final positioning result.

Team name: YAI

Team members: Ying-Ren Chien, Chih-Chieh Yu, Shih-Hau Fang, and Yu Taso

Affiliation: National Ilan University; National Taiwan Normal University; Research Center for Information Technology Innovation, Academia Sinica

Description:
In this technical description, we outline the positioning algorithm and its preliminary results. This study presented a Wi-Fi indoor positioning system that combines fuzzy processing with the smoothed exponential weighted K-nearest neighbors (WKNN) algorithm. The empirical results demonstrate significant improvements in positioning accuracy across various environments and devices. Our positioning algorithm exploits the received signal strength (RSS) values of WiFi receivers, the values of three-axis accelerator meters, and the pressure sensors. By using the validation data, the third quartile of the positioning errors of positioning errors is around 6.09 meters for empirical testing are from the International Conference on Indoor Positioning and Indoor Navigation (IPIN) 2023.

Team name: AUSMLab

Team members: Afnan Ahmad

Affiliation: York University

Description:
The indoor localization system combines WiFi RSSI and IMU data to solve the problem of signal loss and multipath propagation in indoor environments. It uses deep learning techniques, specifically a VAE-LSTM encoder and an Online Feature Adaptive Module (OFAM), to enhance accuracy, adaptability, and resistance to signal noise. The criteria include localization accuracy, adaptability to environmental changes, and signal variation resistance. The system first processes data with noise and motion augmentation before combining WiFi and IMU information to make robust predictions. OFAM provides real-time adaptation to maintain excellent accuracy under changing situations.

Team name: THWS & cronn

Team members: Frank Deinzer

Affiliation: Technical University of Applied Sciences Wuerzburg-Schweinfurt

Description:
The indoor localization system used in the competition employs a particle filter based on the CONDENSATION algorithm. The system utilizes Bluetooth low energy beacons, Wi-Fi, and an inertial measurement unit (IMU) in a commercial smartphone to gather data. The process includes three main stages: transition, evaluation, and resampling. The transition stage updates particle positions using IMU data, while the evaluation stage assesses these positions based on signal strength models for Wi-Fi and Bluetooth. Resampling addresses weight degeneracy by selecting particles based on their weights. The system ensures real-time performance and can handle sensor outages.

Team name: CEPNT

Team members: Baoguo Yu, Qingwu Yi, Jun Li, Yanan Hu, Haonan Jia, Denghui Du

Affiliation: The 54th Research Institute of China Electronics Technology Group

Description:
Multiple conditions zero-velocity detection algorithms are employed to detect zero-velocity intervals within the walking gait. Within the detected zero-velocity intervals, the principle of zero-velocity correction algorithm is utilized to construct the observed quantity of velocity error; leveraging the characteristics of stationary pedestrian foot-ground contact and only subjected to gravity acceleration and unchanged posture angles within the zero-velocity intervals to construct the observed quantity of posture angle error. Satellite signals acquired by the system are differentiated and identified, usable signals are extracted, and a Kalman filter is applied to estimate errors. The obtained error state estimation results are utilized for error correction in pedestrian navigation, enhancing the accuracy of inertial pedestrian navigation.

Team name: Flugmesstechnik für Fußgänger

Team members: Michael Kohl, Benjamin Greiner, Jörg Wagner

Affiliation: Chair of Flight Measuring Technology, University of Stuttgart

Description:
The competing software system represents a ZUPT (Zero Velocity Update) based pedestrian inertial navigation system (INS) with its focus on inertial sensor data without relying on GNSS measurements for aiding. The system uses a total-state, continuous-discrete extended Kalman Filter, which includes inertial sensor bias estimation and employs a Runge-Kutta algorithm of fourth order. The system furthermore includes earth's rotation rate in the mechanization equations and applies gyrocompassing techniques in an initial alignment process within the scope of possibilities of the competition. Within these scopes, additional aiding techniques might be utilised (will show after processing the "Testing Trial")

Team name: X-lab

Team members: Xiaodong Li

Affiliation: Nanjing University of Aeronautics and Astronautics

Description:
Our indoor positioning system employs a multi-faceted approach to achieve precise and reliable results. The foundation of our method is the Zero Velocity Update (ZUPT), which corrects inertial drift by resetting the IMU's velocity to zero during specific motion phases. To further enhance IMU accuracy, we perform Allan variance analysis to correct biases, addressing long-term drifts and noise. We also calibrate the magnetometer to compensate for local magnetic disturbances, improving orientation estimates. For vertical positioning in multi-story environments, we utilize barometric pressure data to detect floor changes. Additionally, our system leverages GNSS signals when available to periodically recalibrate and refine position estimates, ensuring accuracy even in environments where GNSS coverage is limited.

Team name: Warrior Navigation Team

Team members: Sun Wei, Jiang Rong, Zhuang Guangchen

Affiliation: Beijing Institute of Automation Control Equipment

Description:
To address the urgent need for high-precision navigation information for pedestrians in satellite-denied environments, we propose an artificial intelligence-assisted foot-mounted individual soldier navigation and positioning algorithm. This algorithm utilizes micro-electro-mechanical system (MEMS) gyroscopes, MEMS accelerometers, magnetometers, pressure sensors, and other sensors to achieve high-precision autonomous pedestrian navigation and positioning. The proposed navigation and positioning algorithm consists of three main phases: coarse alignment, fine alignment, and navigation.

Team name: InLocate

Team members: Yan Wang, Xuehang Sun, Jian Kuang, Shiyi Chen, You Li, Li Cao, Liping Xu

Affiliation: Wuhan University, 深圳传音控股股份有限公司

Description:
System Overview: Track 6 localization integrates Pedestrian Dead Reckoning (PDR), Vehicle Dead Reckoning (VDR), fingerprint matching, and GNSS for precise indoor and outdoor positioning.
Neural PDR: Uses deep learning on smartphone IMU data for 3D pedestrian tracking.
VDR: Vehicle motion constraints improve VDR accuracy.
Fingerprint Localization: Matches smartphone signals with a pre-collected Wi-Fi, Bluetooth, and magnetic field database.
GNSS: Combines GNSS with inertial navigation and factor graph optimization for enhanced outdoor positioning.
Scenario Recognition: Dynamically switches algorithms based on real-time context.
Workflow: Combines PDR, VDR, and feature matching for indoor positioning, and optimizes GNSS and IMU data for outdoor localization

Team name: City Navigators

Team members: Jingxian Wang, Yue Yu, Duojie Weng, Sheng Bao

Affiliation: The Hong Kong Polytechnic University

Description:
Our system comprises two parts: pedestrian navigation and vehicle navigation. For pedestrian navigation, instead of detecting specific indoor/outdoor positioning scenes and corresponding location sources, we use a transfer learning-based error prediction model. This model provides undifferentiated error evaluation results across various location sources. We extract three different groups of features from these sources for model-based transfer learning and employ a hybrid deep-learning structure for training. For vehicle navigation, the GNSS module, barometer, and IMU embedded in the smartphone supply data to the system-on-a-chip (SoC). Initially, the INS is applied, using raw IMU data as inputs to DeepOdo to estimate the vehicle's forward velocity. GNSS results and barometer-based altitude are then integrated with the INS in an extended Kalman filter (EKF).