The research in video surveillance is moving towards semantic (i.e., meaning) analysis of the scene contents through high-level descriptions. Chaos theory has been reported to simulate partial functions (i.e., neuronal activity in brain) of the Human Visual System. In this work, we propose a chaotic synchronization-based method to identify semantic entities in surveillance scenes. MPEG-7 visual Descriptors(Ds) are used to extract low-level features of video objects. These objects are generated per video frame by segmentation and tracking. The chaotic synchronization is used to perform feature binding (i.e., group semantically relevant feature elements) from these Ds. The objective is to search for unique numeric descriptions (based on low-level features) to identify semantic entities. The idea of a semantic space is introduced to explain feature binding from multiple features spaces. Subjective result evaluation of our results show the existence of such numeric description for related semantic entities (e.g., male, female, enter, deposit, take object, two person meet).