Human action classification distinguishes different human behaviors at a video signal. Suspicious behavior can be defined by the user, and in long distance imaging it may include bending the body during walking or crawling, in contrast to regular walking for instance. When imaging is performed through relatively long distance, some difficulties occur which affect the performances regular action recognition tasks. The degradation sources that include turbulence and aerosols in the atmosphere cause blur and spatiotemporal-varying distortions (image dancing). These effects become more significant as the imaging distance increases and as the sizes of the objects of interest in the image are smaller. The process of action recognition is usually a part of surveillance system that naturally includes a detection of the moving objects as a first step, followed by tracking them in the video sequence. In this study, we first detect and track moving objects in long-distance horizontal imaging, and then we examine dynamic spatio-temporal (motion and shape) characteristics of correctly detected moving objects. According to such characteristics. We construct features that characterize different actions for such imaging conditions, and distinguish suspicious from non-suspicious actions, based on these characteristics.