Gait phase detection and quantitative evaluation are significant for synchronous robotic assistance of human walking, rehabilitation training, or diagnosis of human motion state. Especially, accurate heel-contact detection in a gait cycle is a key requirement for gait analysis applications. Some techniques have been proposed by utilizing wearable devices, however, existing systems typically require precise and continuous time-series data at every single timestep for calibration, which largely increases the burden to users. Therefore, we propose a novel posing-based detection method through measuring muscle deformation, which only requires arbitrary and discrete posture data for calibration without walking. In this study, we firstly collected the posing data as the training set and gait data as the test set from participants through a FirstVR device. Then the Support Vector Machine was trained to be a two-class classifier of heel-contact and non-heel-contact phases by using the collected muscle deformation data during posing. Finally we propose an efficient evaluation system by taking advantage of OpenPose to automatically label our continuous gait data. Experimental results demonstrate the muscle deformation sensor could correctly detect heel-contact with approximately 80% accuracy during walking, which shows the feasibility of posing-based method with muscle deformation information for heel-contact detection.