Estimation of Antenna Pose in the Earth Frame Using Camera and IMU Data from Mobile Phones

Zhen Wang, Bingwen Jin, Weidong GengSensors 2017 · OpenReview

TL;DR

This paper introduces a novel, non-contact method for estimating antenna pose (downtilt and azimuth angles) using only a smartphone's camera and IMU. By combining multi-view images with a 3D antenna model and leveraging improved camera-IMU calibration, the system achieves high accuracy suitable for telecommunication network optimization, avoiding risky and expensive manual measurements.

What Problem Does This Paper Address?

Accurate antenna pose is crucial for optimizing cellular network performance, affecting coverage, interference, and signal quality. Traditional methods involve manual measurements by tower climbers or specialized, expensive equipment. These methods are labor-intensive, risky, and can be costly for large-scale deployments. Furthermore, existing mobile phone-based solutions often struggle with the sparse texture and simple geometry of antennas, and the inherent inaccuracies of low-cost IMU sensors. This paper aims to bridge this gap by providing an accurate, non-contact, and accessible solution using readily available smartphone technology.

Methods

The core of the proposed method lies in a two-stage approach:

  1. Improved Camera-IMU Calibration: The authors address the inherent inaccuracies of mobile phone IMUs by proposing a new calibration method. This method introduces additional degrees of freedom to account for potential misalignments between the accelerometer and magnetometer sensors. Crucially, it uses a novel error metric that considers both downtilt and azimuth errors separately, rather than a single unified rotational error, leading to better simultaneous accuracy for both.
  2. Coarse-to-Fine Visual Pose Estimation: For estimating the antenna's pose relative to the camera, the paper employs a multi-view approach using a known 3D antenna model.
  3. Initialization: They develop an automatic method to find an initial approximate pose by aligning the 3D principal axis of the antenna model with the detected 2D principal axes in the images. This avoids manual initialization.
  4. Refinement: To improve accuracy and avoid local minima, they introduce bounding constraints on the pose search space. This transforms the unconstrained optimization problem into a bounded one, allowing for more robust global optimization.

Finally, the estimated antenna pose in the camera frame is transformed into the Earth frame using the refined camera-IMU calibration and IMU orientation data.

Datasets

The paper utilizes several datasets for evaluation:

  • Camera-IMU Calibration Datasets:
    • BoardF: Multi-view checkerboard data captured in a hall and balcony, with the checkerboard aligned horizontally and northwards. Used for evaluating calibration methods.
  • BoardL: Similar to BoardF but without orientation restrictions, treating checkerboard angles as ground truth. Used for training and evaluating their proposed calibration method.
  • Motion: Video and gyroscope data used for evaluating the CRISP calibration method.
  • Multi-View Antenna Datasets:
    • AntennaL: 65 groups of multi-view antenna images captured on campus, with varying downtilt and azimuth angles. Used for evaluating the overall pipeline.
  • AntennaS: Similar to AntennaL but collected from working antennas, with fewer groups but more viewpoints per group. Used for evaluating performance in a real-world environment.

Key Insights and Ideas

  • Decoupling IMU sensor errors is crucial: The paper highlights that treating accelerometer and magnetometer misalignments as separate degrees of freedom, rather than a single unified rotation, significantly improves azimuth estimation accuracy.
  • Separate downtilt and azimuth error metrics are better: Using distinct error metrics for downtilt and azimuth angles, instead of a single rotational error, leads to more balanced and accurate estimations for both.
  • Constrained optimization is key for robustness: By establishing bounds on the pose search space, the system can avoid local minima and achieve more accurate pose estimations, especially for textureless objects.
  • Smartphone sensors are sufficient (with good calibration): The work demonstrates that with careful camera-IMU calibration and a well-designed visual pose estimation pipeline, standard smartphone sensors can achieve professional-grade accuracy for specific tasks.
  • Contour-based matching is robust for simple shapes: Relying on object contours rather than sparse features makes the method suitable for textureless objects like antennas, which are challenging for traditional feature-based methods.

Key Results

  • Camera-IMU Calibration: The proposed method achieves an average improvement of approximately 1.0 degree in azimuth accuracy compared to existing methods, while maintaining similar downtilt accuracy.
  • Visual Pose Estimation: When integrated into their pipeline, state-of-the-art visual pose estimation methods achieve satisfactory results in over 75% of cases. Their constrained optimization approach further reduces estimation errors by an average of 0.13 degrees for downtilt and 0.3 degrees for azimuth, and significantly in the worst case (0.15 and 7.3 degrees lower, respectively).
  • Overall Pipeline Accuracy: The complete system achieves average downtilt and azimuth accuracies of 0.47 and 5.6 degrees, respectively, with worst-case errors of 1.38 and 12.0 degrees. These results meet the accuracy requirements for telecommunication network optimization.
  • Real-world performance: On working antennas, the largest errors are below the industrial tolerance values (1.5 degrees for downtilt, 15 degrees for azimuth), demonstrating practical applicability.

Limitations and Future Work

  • Computational Cost: The paper acknowledges that the global optimization step in the pose refinement procedure is computationally intensive.
  • Shake Impact: For applications where a tripod is not used, handshakes can affect accelerometer data. While a simple mean filter can mitigate this, more advanced image stabilization techniques are suggested for future work.
  • Fusion of Multi-view Measurements: The current averaging strategy for fusing pose measurements from multiple viewpoints might fail with outliers. Future work could explore more robust fusion methods and potentially integrate information from additional sensors.
  • User Interaction: The authors suggest that simple user interactions or more heuristic search space reduction strategies could alleviate computational burden.
  • Broader Applications: The method's potential utility is noted in fields beyond telecommunications, such as space exploration, indoor navigation, and unmanned aerial vehicles.

Summary updated

Abstract

The poses of base station antennas play an important role in cellular network optimization. Existing methods of pose estimation are based on physical measurements performed either by tower climbers or using additional sensors attached to antennas. In this paper, we present a novel non-contact method of antenna pose measurement based on multi-view images of the antenna and inertial measurement unit (IMU) data captured by a mobile phone. Given a known 3D model of the antenna, we first estimate the antenna pose relative to the phone camera from the multi-view images and then employ the corresponding IMU data to transform the pose from the camera coordinate frame into the Earth coordinate frame. To enhance the resulting accuracy, we improve existing camera-IMU calibration models by introducing additional degrees of freedom between the IMU sensors and defining a new error metric based on both the downtilt and azimuth angles, instead of a unified rotational error metric, to refine the calibration. In comparison with existing camera-IMU calibration methods, our method achieves an improvement in azimuth accuracy of approximately 1.0 degree on average while maintaining the same level of downtilt accuracy. For the pose estimation in the camera coordinate frame, we propose an automatic method of initializing the optimization solver and generating bounding constraints on the resulting pose to achieve better accuracy. With this initialization, state-of-the-art visual pose estimation methods yield satisfactory results in more than 75% of cases when plugged into our pipeline, and our solution, which takes advantage of the constraints, achieves even lower estimation errors on the downtilt and azimuth angles, both on average (0.13 and 0.3 degrees lower, respectively) and in the worst case (0.15 and 7.3 degrees lower, respectively), according to an evaluation conducted on a dataset consisting of 65 groups of data. We show that both of our enhancements contribute to the performance improvement offered by the proposed estimation pipeline, which achieves downtilt and azimuth accuracies of respectively 0.47 and 5.6 degrees on average and 1.38 and 12.0 degrees in the worst case, thereby satisfying the accuracy requirements for network optimization in the telecommunication industry.
BibTeX
@inproceedings{wang2017estimation,
  author = {Zhen Wang and Bingwen Jin and Weidong Geng},
  title = {{Estimation of Antenna Pose in the Earth Frame Using Camera and IMU Data from Mobile Phones}},
  booktitle = {Sensors},
  year = {2017},
  url = {https://openreview.net/forum?id=qVoEfroOO5},
}
Figure