A method and system for capturing and filtering surveillance images are described. A processor detects motion of an object in a field of view and then generates a plurality of images in response to detecting motion of the object in the field of view. A first filter is then applied to the plurality of images and later, the one or remaining images based on the first filter are stored in a memory device. In response to a triggering event, an energy conserving processor is activated from a sleep mode of operation where the energy conserving processor applies a second filter to the remaining images from the first filter. The energy conserving processor assigns a confidence score to one or more images matching the second filter. With the second filter, the energy conserving processor selects fewer images for RF transmission to a remote image analyzer, resulting in increased power savings.