Syntiant Corp. has launched a new machine learning-powered vision system for security cameras that enables ultra-fast and precise identification of specific people or objects.
The system was demonstrated earlier this month by the Irvine-based AI computing company at the ISC West 2025 exposition in Las Vegas.
Potential applications range from police patrolling to vehicle traffic monitoring to surveillance of businesses and smart homes equipped with electronic devices.
One example: a user can say, “Find people wearing yellow jackets,” and the surveillance footage will immediately turn up images of those people – even if the system has never been trained on that specific request.
Search results can also include images, audio and text descriptions of scenes from multiple cameras operating together.
Although the Syntiant technology carries the daunting name of “multimodal vision transformer,” its functionality is straightforward.
The company exclusively provided the Business Journal with an explanation, using a real-world use case:
“If deployed at a shopping mall, an operator could enter a search query like ‘find the tall person with the red hat’ or ‘find the red Toyota with a roof rack’ and the system returns all times and locations where the person or vehicle was seen,” the company told the Business Journal.
“In addition, the search can be done using an image, or a combination of an image and text, so for example, the operator could enter an image of a particular vehicle, and the system would find all the times and locations where the vehicle was seen.”
Finding a Red Backpack
Another example. An operator can type “Find a red backpack,” and the system will show all the times and places the backpack was seen. The operator can also use a picture of a similar backpack, and the system will look for anything that matches.
A key feature of the Syntiant algorithm is that it can run entirely at the “edge”—for example in a security camera—without the delay and extra expense involved in sending information to the cloud and back.
This means that the camera video feed does not need to be transferred out of the camera, and there is no need to have expensive cloud services to process the imagery.
The system shows all times and locations where the person or vehicle was seen, though some people may find it unsettling to know just how easy and fast it is to locate them in a crowd or other public space
A further explanation from the company: “Think of it like this: if you wanted to know what your dog did all day, you could either watch hours of video or just ask this system, ‘Show me every time my dog chased its tail.’ Instead of sending all the video to the internet to be processed, the camera itself can answer your question.”
Syntiant is working with a variety of potential customers.
The company was founded in 2017. Since then, more than 100 million of its chip sets and machine learning models have been deployed worldwide, along with billions of MEMS ultra-tiny microphones and sensors.
Syntiant’s products can also detect audio-based events including glass breaking, smoke alarms and carbon monoxide alarms.
