Gait recognition is an effective biometric authentication system that can be used for the identification of individuals at a distance, when other methods prove to be ineffective. Most of the research in this field has been visible spectrum-oriented and sensor based, with only a few attempts at exploring recognition in the thermal infrared band. This paper proposes a non-deep learning based approach for gait recognition under poor lighting conditions via thermal infrared imagery by investigating the basic concept of Point Light Animation (PLA) for gait representation. Firstly, background subtraction is performed on each frame of the gait sequence followed by removal of silhouette noise with the help of morphological operations. The gait sequences are then distinguished by the relative motion of various body points along with spatio-temporal features that are fused at a feature level for classification using Support Vector Machines. The proposed algorithm is evaluated on the CASIA Dataset-C and achieves a peak accuracy of 85.78% with all the features used for classification.