This thesis presents a system for adaptive soft object grasping using visual deformation tracking. The goal was to enable a robotic gripper to adjust its behavior based on how an object deforms during compression, using low-cost and interpretable sensing. A custom setup was developed to collect synchronized RGB-D and force data during gripping, allowing stress and strain to be calculated and used to estimate physical properties such as Young's modulus and Poisson's ratio. Three calibration methods were explored: vision-only, force-only, and a combined approach. Although absolute estimates of mechanical properties did not align with expected reference values, the system reliably distinguished between materials based on relative deformation patterns. The most accurate stiffness estimation was achieved using a single-point method where gripper width and force were held constant, rather than from regression over the full grasping cycle. Based on this calibration, object-specific deformation thresholds were assigned and used to control the gripper in real time. A visual tracking script monitored the strain, while a separate control node closed the gripper until the threshold was reached. This approach was tested in a mock scenario using water-filled sponges, where it reduced water loss to 6.88% compared to 23.28% with the gripper's default mode for soft objects. These results demonstrate that vision-based deformation tracking can support practical and adaptive control without relying on complex models or expensive sensors. The findings suggest that even simple visual feedback, when integrated into a lightweight control strategy, can improve soft object handling in robotics.