An ROI Optimization Method Based on Dynamic Estimation Adjustment Model

Abstract

An important research direction in the field of traffic light recognition of autonomous systems is to accurately obtain the region of interest (ROI) of the image through the multi-sensor assisted method. Dynamic evaluation of the performance of the multi-sensor (GNSS, IMU, and odometer) fusion positioning system to obtain the optimum size of the ROI is essential for further improvement of recognition accuracy. In this paper, we propose a dynamic estimation adjustment (DEA) model construction method to optimize the ROI. First, according to the residual variance of the integrated navigation system and the vehicle velocity, we divide the innovation into an approximate Gaussian fitting region (AGFR) and a Gaussian convergence region (GCR) and estimate them using variational Bayesian gated recurrent unit (VBGRU) networks and a Gaussian mixture model (GMM), respectively, to obtain the GNSS measurement uncertainty. Then, the relationship between the GNSS measurement uncertainty and the multi-sensor aided ROI acquisition error is deduced and analyzed in detail. Further, we build a dynamic estimation adjustment model to convert the innovation of the multi-sensor integrated navigation system into the optimal ROI size of the traffic lights online. Finally, we use the YOLOv4 model to detect and recognize the traffic lights in the ROI. Based on laboratory simulation and real road tests, we verify the performance of the DEA model. The experimental results show that the proposed algorithm is more suitable for the application of autonomous vehicles in complex urban road scenarios than the existing achievements.

Resilient Factor Graph-Based GNSS/IMU/Vision/Odo Integrated Navigation Scheme Enhanced by Noise Approximate Gaussian Estimation in Challenging Environments

“…In the three sets of data, PDF1, PDF2, and PDF3, the RM set to 0.3 m/s, and the distribution characteristics gradually evolved toward the Ga distribution as the velocity increased. At the same time, in the two sets of data, PD PDF4, the velocity value was 3.754 m/s, and when the RMSE was reduced from 0.3 0.03 m/s, the distribution characteristics became closer to the Gaussian distributio shows that as the error-to-true value ratio decreases, the measurement noise error will gradually evolve from a Rayleigh distribution to an approximate Gaussian distri In our previous research [34], using a simulation platform, we conducted a d analysis of the changes in the velocity and variance function relationship curves Based on the vehicle speed and innovation data distribution characteristics in 2, a convergence function 𝑦 = 𝑓 (𝑥) is fitted, and the data regions satisfying the r ship, 𝑦 ≥ 𝑓 (𝑥) and 𝑦 𝑓 (𝑥), are divided into the AGFR and the AGCR, respecti the AGFR, the nonlinearity of the speed and variance relationship function is re strong, and a neural network is suitable for fitting and solving. In the AGCR, th mation sequence has converged, and it can be estimated according to the Gaussian bution.…”

“…Figure 1. The residual sequence distribution characteristics.In our previous research[34], using a simulation platform, we conducted a detailed analysis of the changes in the velocity and variance function relationship curves during the uniform acceleration of a vehicle from [0 m/s, 0 m/s] to [6 m/s, 3 m/s] for nine sets of GNSS velocity RMSE values. The results are shown in Figure2.…”