The current security surveillance system offers two modes for reviewing video recordings: blind search and image search.
Video Blind Audition, Inefficient and Time-Consuming:Unable to determine the specific location or time frame of the footage being searched, we can only review the target footage frame by frame, which is inefficient and may not yield the desired content.
First Image, Then Search:The model is suitable for retrieving the locations visited by a person through facial or body image searches. A target person's sample image is required, and it does not support searches for other objects.
Intelligent Text Search
Hikvision launches an intelligent text search solution, based on AI large model technology, which identifies various objects in video surveillance images, establishes a database of such objects, and then combines with AI agents to achieveNatural Language SearchTarget Video. Simplifies the process of reviewing surveillance footage, significantly enhancing the efficiency of video retrieval.
Intelligent Text Search allows users to retrieve targeted video clips by inputting text, such as entering "a man wearing a white hat." The system then searches for video recordings and snapshots within a specified time frame that feature a man wearing a white hat. By inputting "boxes," users can query all video clips containing boxes from the surveillance footage.
Garden Complex Building ImplementationBy searching for cardboard boxes, investigate incidents of fire lane occupation; by searching for security personnel in the park, review the patrol trails of individual guards; by searching for electric scooters, check for violations of illegal parking in designated no-parking zones...
Branch Network ImplementationThrough cash boxes/courier vehicles, review compliance and timeliness of handover processes at sites; identify recent renovation sites for tool recognition: hard hats, reflective vests; by searching for security guards, review security guards' scheduled patrols of ATMs...
Search Anything
Traditional AI intelligent recognition focuses on a specific type of object, requiring the pre-set AI model, followed by training and deployment, yet it cannot achieve universal recognition.
Hikvision's current text search AI model supports the recognition of over 6,000 items. Compared to traditional AI models that identify single items, our model offers a one-time setup with more comprehensive data, more flexible video retrieval, and a broader range of application scenarios.
During the use of SmartText Search, when encountering examples of incorrect search results, users can manually classify the data as correct or incorrect. The system will quickly fine-tune based on the classified examples. Subsequent searches will significantly improve accuracy, achieving the effect of becoming more precise with use.
Currently, large-scale language model products in the market are still based on high-performance GPUs.Server ModeHigh cost of use? Hikvision integrates multimodal large models with embedded hardware, achieving lightweight edge-end solutions.NVRSeries of products. Achieving a balance between performance and cost, enabling the use of large-scale language search applications at more affordable prices.
In 2024, Hikvision passed the可信AI Large Model Standard Compliance Verification organized by the China Information and Communication Research Institute: the model development received a 5-star rating, and the model capability was rated at 4+. There are no more than 4 companies in China with model development and capability reaching level 4 and above.
