Soft robots enable safe and adaptive interaction with their environment, but still face major challenges in detecting contact during manipulation. This paper presents a low-cost vision-based setup designed to obtain groundtruth contact areas of soft manipulators without requiring the embedding of sensors into their structure. The platform uses an external camera and controlled lighting to capture the contact interface under different materials and configurations. Three segmentation methods, HSV thresholding, the Segment Anything Model (SAM 2.1), and a CNN with a VGG16 encoder, were compared for performance. Among them, optimized HSV thresholding achieved the best balance between accuracy and simplicity, with a Dice score of 0.97 and a mean error of 8% on reference tests. The proposed setup provides a practical and reproducible method to study contact formation in soft robotics and a reproducible method for obtaining ground-truth data for tactile sensing and control. ... Read more