Accurate dietary assessment is crucial for addressing obesity and non-communicable diseases. Traditional methods, such as food diaries and 24-h recalls, often suffer from 20–50% error and lack cultural specificity. This study introduces a stereo-vision pipeline that leverages depth information to estimate food volume with 92–95% accuracy, significantly outperforming 2D methods, which exhibit 15–25% error. A reference cube is used to calibrate scale. The system was evaluated on five traditional Tunisian dishes—couscous, brik, lablabi, tajine, and salad—demonstrating its effectiveness, with mean volume errors consistently below 5%. For example, couscous volume was estimated at 285 cm3 compared to the true 300 cm3 (5% error), while brik showed a 3.3% error. This culturally sensitive approach provides a high-precision tool for dietary monitoring, with potential applications in public health and sports nutrition in Tunisia and beyond.