Location-based services (LBS) have significantly enhanced the convenience of mobile services for the general public, becoming an essential component of our daily lives and work. Accomplishing precise indoor positioning and implementing crowd-sourced mapping within multi-level parking structures presents a significant challenge. Indoor magnetic feature matching is a feasible solution but has to reduce the mapping cost by crowd-sourcing. To address the issue of crowd-sourced magnetic mapping in multi-level garages, we propose a novel solution for constructing crowd-sourced magnetic grid maps based on pedestrian-vehicle collaboration. This approach leverages the high-precision capabilities of Vehicle Dead Reckoning (VDR) to gather magnetic field data and floor information from the primary routes within parking structures, utilizing crowdsourced vehicle data. In processing vehicle crowd-sourced data, we employ a method akin to pedestrian trajectory processing, coupled with floor transition detection, to impose floor-level constraints. Through global trajectory optimization, we achieve precise estimation of 3D vehicle trajectories from the crowdsourced data. Furthermore, by iteratively refining the graph optimization problem, incorporating keyframe association details, floor information, and adjacent loop closure data, we successfully reconstruct the vehicle's 3D trajectory. Field tests demonstrate that the proposed method can achieve an average planar positioning error of 2.19 meters using only vehicle-mounted smartphone sensor data. The relative height error between two floors is a mere 0.08 meters.

Indoor positioning technology is one of the important supporting technologies for location services. Since WiFibased indoor positioning technology is unsuitable for scenarios where WiFi APs are sparse or missing (such as underground parking lots), magnetic fields are the only available signal source. However, magnetic positioning methods still face the problem of the high cost of magnetic map construction. This paper proposes a method for constructing a magnetic grid map of a multi-story parking lot that integrates crowdsourced vehicle and pedestrian data. This method uses the three-dimensional sparse magnetic field map obtained from vehicle data as the skeleton, performs preliminary floor allocation on pedestrian data, and combines key frame association information, floor information, and neighborhood loop information to iteratively update the graph optimization problem to achieve three-dimensional trajectory reconstruction of pedestrians and vehicles. On this basis, this method defines crowdsourced trajectory estimation and magnetic grid map generation as an optimization problem, establishes a connection between the two through multiple constraints such as trajectory relative pose constraints, magnetic field vector space consistency, and magnetic field spatial distribution continuity, and obtains the optimal estimate of magnetic grid maps and trajectories on each floor through a multi-step iterative optimization method. The effectiveness of the proposed scheme is verified by testing a simulated data set in a typical twostory underground parking lot scenario. The average positioning error of the jointly generated trajectories of pedestrians and vehicles is 1.75m, and the average positioning error based on the crowdsourced magnetic field map is 2.87m.

The Internet of Things and location-based services rely heavily on indoor pedestrian navigation. It is important yet costly to create a map for indoor navigation. Recent studies examine the creation of radio maps using crowdsourcing (such as WiFi and BLE). In situations when there are insufficient WiFi or BLE nodes, the magnetic field map is the sole way to deliver indoor locating services without infrastructure. Previous research on magnetic map construction using crowdsourcing do not fulfill the requirements of the real scenario. In this work, we provide a summary of the real-world requirements for the crowdsourcing- based magnetic map-building method. Furthermore, we proposed an efficient crowdsourcing-based magnetic map creation approach that is resilient to the proposed requirements: short- period trajectories, uncalibrated magnetometer sensors, different pedestrian motion patterns, and irregular trajectory shapes. In the field test, the proposed method consumed 60.8 seconds to generate a map from a 12-hour dataset that met the proposed requirements. When the duration of each trajectory is 90 seconds, the mean position error is 1.48m (with scale correction) and 2.53m (without scale correction).

KITTI dataset is collected from three types of environments, i.e., country, urban and highway The types of feature point cover a variety of scenes. The KITTI dataset provides 22 sequences of LiDAR data. 11 sequences of them from sequence 00 to sequence 10 are “training” data. The training data are provided with ground truth translation and rotation. In addition, field experiment data is collected by low-resolution LiDAR, VLP-16 in Wuhan Research and Innovation Center.

The 3D position estimation of pedestrians is a vital problem in the development of virtual reality, augmented reality, and the internet of things. The learning-based inertial odometry is a very potential auxiliary method of pedestrian positioning due to its low position drift and immunity to external environmental influences. However, in many cases, the drift error of the heading is still the main factor that causes the rapid divergence of the position estimated by the learning based inertial odometry. This paper proposed a graph optimization-based estimation method to fusing learned based inertial odometry and magnetometer measurements for obtaining lower drift position. The proposed algorithm does not need to calibrate the magnetometer bias, and effectively resist the influence of magnetic interference in the indoor environment, and can provide a very reliable absolute magnetic heading correction. Test results show that the proposed method can obtain better positioning performance than other methods using calibrated magnetometer observations.

The 3D position estimation of pedestrians is a vital module to build the connections between persons and things. The traditional gait model-based methods cannot fulfill the various motion patterns. And the various data-driven-based inertial odometry solutions focus on the 2D trajectory estimation on the ground plane, which is not suitable for AR applications. TLIO (Tight Learned Inertial Odometry) proposed an inertial-based 3D motion estimator that achieves very low position drift by using the raw IMU measurements and the displacement predict coming from a neural network to provide low drift pedestrian dead reckoning. However, TLIO is unsuitable for mobile devices because it is computationally expensive. In this paper, a lightweight learned inertial odometry network (LLIO-Net) is designed for mobile devices. By replacing the network in TLIO with the LLIO-Net, the proposed system shows similar accuracy but significantly improved efficiency. Specifically, the proposed LLIO algorithm was implemented on mobile devices and compared the efficiency with TLIO. The inference efficiency of the proposed system is 2-12 times that of TLIO.