In the SPI, 99% of the data is unstructured. The improvement of intelligent security system has been limited for a long time because of the stagnation in the development of traditional algorithms and chip technologies. Therefore, Deep Learning is required to make breakthroughs, so that objects can be studied, scenes segmented, people and vehicles analysed.
On the 2016 China International Exhibition on Public Safety and Security, Deep Learning was seen as the lynchpin for breakthroughs. The advancement in technology enables the SPI to develop more quickly and extend into more fields, and grants video more value.
In terms of “developing more quickly”, companies used to focus on “seeing clearly” (high definition video images and the equipment’s capability to work in harsh environments), but now they attach more importance on helping users to “understand what they see”. In other words, the former emphasises the evidence collection after incidents while the latter strengthens the prevention of such incidents. Surmounting the difficulty of unstructured videos, the monitoring system is able to study and judge the situation, and dynamically recognize human faces, which remarkably promotes the real-time alarm rate and reduces the false alarm rate.
In terms of “extending into more fields”, the video monitoring system not only meets the demand for security and protection but also further explores the value of videos by virtue of Deep Learning and the computing capacity of unstructured data, such as making business decisions and improving business operational efficiency.
As a matter of fact, the SPI can draw lessons from the experience in other fields. State-of-art image chip technologies and algorithms have been widely used in areas such as dynamic facial recognition, intelligent transportation system, unmanned aerial vehicle(UAV) image processing and remote sensing geographical image processing.