This article focuses on the practical application scenarios of VSLAM (Visual Simultaneous Localization and Mapping) and VIO (Visual-Inertial Odometry). Combining the characteristics of global shutter cameras and odometers, it shares actionable optimization tips for these two key components, providing references for engineering implementation.

I. Outdoor Application and Optimization Tips for Global Shutter Cameras

1. Rationale for Prioritizing Global Shutter Cameras in VIO

For VIO systems, global shutter cameras are recommended as the preferred choice for outdoor scenarios. Although rolling shutter cameras can be barely adapted in a technical sense, their exposure time dynamically adjusts based on factors such as ambient light intensity and scene texture due to their working principle. This easily causes issues like motion blur and frame timing misalignment, reducing the stability of VIO’s pose estimation. Moreover, troubleshooting such problems is highly challenging, making rolling shutter cameras unable to meet the reliability requirements of engineering applications.

Before event cameras are commercially deployed on a large scale, global shutter cameras remain the optimal option for VIO visual sensors. Additionally, for nighttime navigation and imaging needs, long-wave infrared (thermal imaging) technology will remain an unrivaled solution in terms of performance for the next decade. From a long-term perspective, if technological evolution aligns with current analysis, the primary sensors in the autonomous driving field may form a competitive landscape between “LiDAR” and “thermal imaging cameras”.

2. Practical Application Pain Points and Optimization Tips for Global Shutter Cameras

Currently, global shutter cameras generally face the issue of high costs. Therefore, most engineering solutions opt for models that output monochrome Y-channel data, with lens aperture specifications often being entry-level parameters such as f/3.0. This type of configuration performs adequately in indoor scenarios with stable texture and lighting. However, in outdoor environments with dynamic light conditions, it is prone to reduced accuracy in VIO frontend feature extraction and matching due to photometric fluctuations, thereby affecting the overall system performance.

If a 1080P resolution RGB global shutter camera (supporting auto-exposure and auto-gain) is adopted, the aforementioned issues can be avoided. Such high-end configurations are not within the scope of this basic tips discussion. For mainstream low-cost monochrome global shutter cameras (with 8-bit Y-channel data output), their outdoor adaptability can be optimized through the following method:

Core Tip: In the VIO system, sample the RAW data output by the camera at a fixed frame rate (e.g., 5Hz) and calculate the pixel mean value of each frame of RAW data. This operation incurs extremely low computing overhead (less than 1% of a single core’s utilization). Based on this mean value, a “custom adaptive photometric adjustment strategy” can be developed for VIO. By dynamically adjusting camera parameters, the photometric value is stably maintained within the ideal working range of indoor scenarios for a long time. This reduces the interference of outdoor light intensity fluctuations on VIO feature tracking and state estimation, helping avoid debugging detours caused by photometric adaptation issues.

II. Application Tips for Odometers in VIO Systems

1. Core Positioning of Odometers: Not a “Necessity for Tight Coupling”

Odometers are often used as a performance benchmark (Benchmark) for VIO systems, and many researchers focus on “how to tightly couple odometers with VIO”. However, from the perspective of engineering practice, tight coupling of odometers is not the optimal solution. On one hand, tight coupling requires additional design of data fusion logic, increasing system complexity; on the other hand, it significantly raises computing overhead, which contradicts the engineering requirements of VIO for “low overhead and high real-time performance”, making it an over-optimized “involutionary” approach.

Some solutions consider “constructing a DR (Dead Reckoning) system using odometer + IMU Kalman filtering to constrain VIO errors”. However, this design also has redundancy — VIO itself already integrates IMU data, and adding an additional DR system increases hardware costs and debugging difficulty, resulting in low cost-effectiveness.

2. Efficient Application Solution: Loose Coupling Verification Between Odometer and VIO

A core characteristic of odometers is that “odometry counts tend to be overestimated” (e.g., wheel slipping and idling can lead to inflated odometry data). Based on this characteristic, a loose coupling verification mode is recommended:

1. Rely on VIO in Normal Operation: The pose estimation reliability of a VIO system with qualified performance can reach over 95%. During daily operation, the position and motion data output by VIO serve as the core basis.
2. Regular Data Verification: Compare the translation distance (e.g., x/y-axis motion odometry) output by VIO with odometer data at fixed intervals (e.g., 250ms). If there is a significant deviation between the two (e.g., VIO indicates a movement of 3.5 meters while the odometer only records 2 meters), it suggests that the VIO may have scale inaccuracies or state drift issues.
3. Rapid Error Correction: When a significant deviation is detected, first trace back historical data for normalization calibration. If the problem persists after calibration, re-execute VIO initialization at the current position and only fine-tune the x/y-axis translation parameters in the 6 Degrees of Freedom (6DoF) to quickly restore system accuracy.

This solution does not require complex fusion algorithms. Through only lightweight verification and correction, it can fully leverage the error monitoring value of the odometer while avoiding additional computing overhead, which aligns with the core engineering requirements of “high efficiency, reliability, and low cost”.