Embedded vision systems are essential for contemporary applications, including robotics, advanced driver assistance systems (ADAS), and intelligent surveillance; yet they frequently experience diminished image quality due to resource constraints, environmental variability, and inconsistent illumination conditions. Such degradations impact multiple visual attributes-sharpness, contrast, color accuracy, noise levels, and structural similarity-that are critical for reliable perception in safety- and performance-driven domains. This study introduces a comprehensive system-level calibration architecture that integrates three coordinated layers: sensor-level adjustment, firmware optimization, and adaptive software enhancements. At the sensor level, exposure control, gain tuning, and white balance adjustments mitigate luminance imbalance and color shifts under changing light conditions. Firmware optimization leverages image signal processor (ISP) parameters to reduce temporal and spatial noise, refine tone mapping, and correct color reproduction through calibrated color correction matrices. Software-level improvements apply adaptive sharpening, contrast enhancement, and gamma correction to maintain visual fidelity across diverse scenes. The proposed pipeline was evaluated on three representative embedded platforms-NVIDIA Jetson Nano, Raspberry Pi 4B, and STM32F7 MCU-covering a range of computational capabilities and power budgets. Experimental results demonstrate substantial improvements in image quality: Peak Signal-to-Noise Ratio (PSNR) increased from 24.2 dB to 31.6 dB in indoor low-light conditions, Structural Similarity Index (SSIM) improved from 0.73 to 0.88 in dynamic scenarios, and color accuracy (ΔE) was reduced to 3.1 in bright outdoor conditions. The complete calibration pipeline sustained real-time responsiveness (<40 ms/frame) with acceptable power consumption (maximum 172 mW) and memory utilization (peak 35.7 MB). These results validate the modularity, efficiency, and robustness of the proposed method, making it well-suited for deployment in practical embedded vision applications where image quality, latency, and resource constraints must be balanced.