INTEGRATION OF ADAPTIVE KALMAN FILTER AND BAYESIAN MODELS FOR DIAGNOSTICS AND RELIABILITY ASSESSMENT OF VISION SENSORS

Vision-based sensors constitute a cornerstone of contemporary automation, robotics, and intelligent transportation systems, where their reliability directly determines operational safety and efficiency. Conventional Kalman filtering achieves optimal state estimation under linear dynamics and stationary noise assumptions, yet its performance deteriorates in the presence of sensor malfunctions or nonstationary environments. This study proposes an innovative framework that couples an adaptive Kalman filter with Bayesian inference to strengthen diagnostic capabilities and reliability evaluation of vision sensors. The Kalman filter provides dynamic state tracking, while Bayesian reasoning enables probabilistic fault identification and reliability quantification. The methodological novelty lies in the real-time adjustment of covariance matrices within the Kalman filter, guided by posterior probabilities derived from Bayesian analysis, thereby ensuring rapid adaptation to sensor anomalies. Analytical results establish bounded estimation error variance and robustness, and simulation experiments demonstrate a marked improvement in accuracy compared to the classical Kalman filter under sensor fault scenarios. The proposed approach exhibits strong applicability in autonomous robotics, unmanned vehicles, and industrial quality assurance systems.