AI-Driven Automatic Side Mirror Adjustment to Address Left-Side Blind Spots in Right Turns

The limited field of view in conventional side mirrors of vehicles poses serious safety risks, particularly during right turns or highway merges where the left-side rear area is often obscured. While digital side mirrors (DSM) offer partial improvements, they fall short in dynamically addressing blind spots in real time. This paper presents an intelligent side mirror control system that automatically adjusts the left-side mirror angle based on real-time driving context. To meet the dual demands of high accuracy and low latency, the system employs a lightweight convolutional neural network (CNN) with a BranchyNet-inspired multi-exit architecture. By enabling early exits under high-confidence predictions using entropy-based criteria, the model significantly reduces inference latency while maintaining reliable performance. Experiments on a custom real-world driving dataset demonstrate that the proposed method achieves up to 97.4% classification accuracy with an average inference time of 1.7 ms, making it suitable for embedded deployment in driver-assistance systems.