Continuous Integration and Continuous Deployment (CI/CD) Frameworks for Safety-Critical Automotive Perception: Automated Drift Detection and Model Refinement in Real-Time Deep Learning Networks

• Abey city

  Lautech

  Author

The integration of Deep Neural Networks (DNNs) into safety-critical automotive perception systems has introduced significant challenges in maintaining model reliability under evolving real-world conditions. While advances in real-time object detection using Convolutional Neural Networks (CNNs) have demonstrated operational viability exceeding 10 Hz on embedded GPU platforms, a critical research gap persists in the continuous validation and automated refinement of these models post-deployment . This study addresses this gap by proposing a comprehensive CI/CD framework that integrates automated drift detection mechanisms with continuous model refinement pipelines for automotive perception systems. The framework employs a hybrid monitoring architecture combining unsupervised driving-state segmentation—achieving 96.8% drift detection accuracy through combined vehicle-kinematic and driver-behavior feature analysis —with supervised performance validation using annotated edge-case datasets. Through retrospective analysis of real-world driving data and prospective simulation of continuous integration workflows, we demonstrate that the proposed framework can detect perceptual performance degradation with 89.4% precision and trigger automated model refinement cycles within operational latency constraints of < 200ms per inference batch. The framework's alignment with emerging ISO 26262 Edition 3 requirements for AI-driven vehicles establishes a replicable methodology for maintaining safety evidence under continuous software updates. This research contributes a systematic approach to operationalizing MLOps principles within the constraints of automotive functional safety standards, providing both theoretical foundations and practical implementation guidelines for the next generation of autonomous driving systems.

• pdf

Copyright (c) 2026 Abey city (Author)

This work is licensed under a Creative Commons Attribution 4.0 International License.