Design and Experimental Validation of an Embedded Edge Artificial Intelligence Camera System for Real-Time Intelligent Vision

Abstract

This study presents the design and experimental validation of an embedded edge artificial intelligence (AI) camera system for intelligent vision applications. The developed system integrates a multicore edge computing processor and a dedicated AI engine into a compact smart camera platform, enabling on-device image processing, intelligent recognition, and localized decision-making. The hardware implementation incorporates circuit design, impedance-controlled high-speed signal routing, and printed circuit board (PCB) stack-up design. The signal integrity of the MIPI Display Serial Interface (MIPI-DSI) is evaluated through scattering-parameter simulations and measurements, while the conducted radio frequency (RF) transmitter performance and radiated emissions of the fabricated prototype are experimentally assessed. In an outdoor and indoor parking-lot field test, the integrated system processed images at approximately 2 frames per second (FPS) while operating in multi-task mode at an image resolution of 1280 × 960 pixels. When configured exclusively for license plate recognition, the system achieves a processing throughput of up to 45 FPS. For all 13 tested vehicles, comprising two buses, two trucks, and nine cars, both the predicted vehicle type and complete license plate number matched the ground truth, yielding an integrated recognition accuracy of 12/13 for this test sample. Representative measurements of the prototype’s DC input power were approximately 2.20–6.32 W at a supply voltage of 12 V. These results demonstrate the feasibility of the developed edge AI camera for fixed-location parking-management applications and real-time edge inference capability.