Detection of Parkinson’s disease (PD) at an early stage is necessary for its treatment. The commonly used methods available in the literature use observation of certain symptoms such as Tremor, Loss of Smell and Troubled Sleeping, Moving or Walking. The motion pattern in this disease can be characterized by a spatio-temporal phenomenon that signifies gait recognition as reported in the literature. However, non-invasive methods such as use of Gait image sequences are handy in terms of cost and comfort. In this paper we propose a statistical approach for detection of Parkinson’s diseases by considering segmental feature of gait image sequences by using Hidden Markov Model (HMM). A set of key features from the image frames is identified during the gait cycle. The input binary silhouette images are preprocessed by morphological operations to fill the holes and remove noise. An image feature vector is created from the outer contour of the image sequences. From the feature vectors of the gait cycle, a set of initial exemplars is constructed. The similarity between the feature vector and the exemplar is measured by the inner product distance. An HMM is trained iteratively using the Viterbi algorithm and Baum-Welch algorithm and then used for detection of Parkisonian gait. The characteristics of one dimensional HMM best fit to one dimensional image vector thus the proposed method reduces image feature from the two-dimensional plane to a one-dimensional vector. The statistical nature of the HMM makes it robust to PD gait representation and recognition. The proposed HMM-based method in LabVIEW and MATLAB is evaluated using the CMU MoBo database as well as our own prepared database for PD detection.