Perception-Based Feedback Tracking via Scalarized Pixel Data of Visual Frames

Perception-based control has gained increasing attention due to its potential in autonomous systems. However, effectively integrating high-dimensional data from perception sensors, such as RGB cameras, into control design remains challenging. This paper investigates a car-following longitudinal control problem in autonomous driving, where a following car equipped with an RGB camera senses visual information from a leading car. To address the complexities of data integration, we propose a Binarized Digital Image Pixels Mask (BDIPM) method integrated with a backstepping control framework. The BDIPM method enhances useful visual pixels while filtering out irrelevant signals, enabling the scalarization of high-dimensional image data into a format suitable for control design with rigorous analysis footing. By constructing an error model that maps image-derived features to real-state variables, we establish a sensory pixel-based feedback control law. CARLA results demonstrate the effectiveness of the proposed approach. Compared to the baseline methods, our algorithm shows improvement in both stability and computational speed, which highlights its potential for robust perception-driven control in autonomous driving scenarios.