Visual inertial odometry (VIO) schemes based on tightly coupled optimization have become a research hotspot in the field of micro unmanned aerial vehicles and autonomous robots. However, the high-precision operation of VIO systems, such as VINS Mono and ORB-SLAM3, depends heavily on accurate initial state estimation, which involves estimating the visual scale factor, gravity direction, initial velocity, and Inertial Measurement Unit (IMU) gyroscope and accelerometer biases during the start-up phase. In practical application scenarios, such as the smooth low-speed start-up of ground robots, sufficient motion excitation is often unavailable, resulting in slow convergence or even divergence of iterative algorithms. To address the issues of heavy dependence on motion excitation, slow convergence, and high computational load during the visual odometry initialization phase of ORB-SLAM3, we propose an improved IMU initialization algorithm based on analytical solutions. Within the ORB-SLAM3 framework, the proposed algorithm replaces the original iterative optimization process with a non-iterative strategy. By constructing polynomial equations for gravity, scale, and accelerometer bias, the optimal estimates of system parameters are directly obtained through root solving. This eliminates the sensitivity of iterative algorithms to initial guesses and the risk of divergence. It mainly comprises two core steps: (1) using rotation constraints to independently estimate and scale-decouple gyroscope bias and (2) constructing a constrained linear system to analytically solve for the remaining inertial parameters. In addition, this method only bears a small part of the computational burden of the original algorithm, saving computing resources and substantially improving the IMU initialization speed. Simultaneously, owing to the high robustness of the proposed algorithm without the need for good initial values, it optimizes the motion excitation threshold, effectively solving the problem of insufficient observability under weak motion conditions. Experimental results on the EuRoC dataset show that the proposed algorithm maintains comparable positioning accuracy to the original ORB-SLAM3 algorithm and outperforms the VINS Fusion algorithm while considerably reducing the requirement for motion excitation. Particularly in low-speed and steady-motion scenarios, the IMU initialization time is reduced by approximately 40%, effectively solving the problems of long initialization time and divergence while substantially improving the system’s real-time performance and robustness. The results demonstrate that the proposed algorithm has high precision, high efficiency, and strong robustness, considerably enhancing the engineering practicality of visual inertial systems in weak motion excitation environments.
- Article type
- Year
- Co-author
Open Access
Issue
To reduce the cost and improve the accuracy of the depth estimation model in the process of image depth information prediction and apply the depth estimation model to the complex football sports scene, an unsupervised monocular image depth prediction method based on the improved FeatDepth is proposed. The monocular depth estimation model is used to obtain the relative depth information between the people, the football, and the goal in the football scene and calculate the distance information between the targets, which can be used for football auxiliary training and monitoring whether the player is offside and other application scenarios.
First, the attention mechanism was introduced to the original FeatDepth method so that the model pays more attention to the effective feature information. Second, the PoseNet and DepthNet networks in FeatDepth were embedded in the GAM global attention mechanism module, adding additional context information to the network and improving the depth prediction performance of the FeatDepth model without increasing the computational cost. Third, because of the higher requirements for depth information prediction in the football scene, to ensure that the model exhibits better performance in the low-texture areas and details, the loss function scheme used by the original FeatDepth method was adopted. The final loss function was mainly composed of the combination of single-view and cross-view reconstruction losses, in which the single-view reconstruction loss was composed of discriminant and convergence losses based on the reconstruction loss, and the cross-view reconstruction loss was composed of eigenmetric and photometric losses. Then, the dataset was made, and the parts of the KITTI public dataset with more person scenes were selected for dataset making, including 4,721 images in the training set, 631 images in the verification set, and 584 images in the test set. Model comparison experiments were conducted to verify the effectiveness of the improvement strategy.
The improved model with the GAM global attention mechanism module is called G-FeatDepth, and after comparative experiments on the dataset, the improved G-FeatDepth model is compared with the improved FeatDepth model on each evaluation index, with absolute relative error decreased by 0.007, square relative error decreased by 0.051, root-mean-square error decreased by 0.032, and root-mean-square logarithmic error decreased by 0.005, as well as accuracy with thresholds δ < 1.25 improved by 0.009, δ < 1.252 improved by 0.004, and δ < 1.253 improved by 0.002, which not only reduces the error index but also improves the accuracy. The experimental data verifies the improvement of the model performance. According to the actual inference effect of the model in the dataset, the G-FeatDepth model has a better depth prediction effect in low-texture areas and details than the other models.
Using the image data in the football scene and comparing the inference effects of each model in the football scene, the improved model G-FeatDepth has a better depth prediction effect in the details of low-texture areas (e.g., football, goals, and limbs), that is, it is more satisfying to predict the depth information in the football scene, and the unsupervised monocular depth estimation model is applied to the football sports scene.
京公网安备11010802044758号