With recent technological developments, criminal investigation has witnessed a revolutionary change in identifying crimes. This has empowered Law Enforcement Agencies (LEAs) to take benefit of such revolution and build a smart criminal investigation ecosystem. Generally, LEAs collect data through surveillance systems (e.g., cameras); which are implemented on public places in order to recognize people behaviors and visually identify those who may form any danger or risk. In this paper, we focus on knives-related crimes or attacks that have been increased in recent years. In order to ensure public safety, it is crucial to detect such type of attacks in an accurate and efficient way in order to help LEAs in reducing potential consequences. We propose a smart video surveillance system (SVSS), which is based on a modified Single Shot Detector (SSD) and is combined with InceptionV2 and MobileNetV2 models. The proposed system is believed to enable LEAs to analyze big data collected from sensor cameras in a real-time and to accurately detect knives-based attacks. Experimental result show that SVSS can achieve better results in real-life scenario in terms of obtaining rapid and accurate attack warnings.