Qualcomm’s Reference Design Enables Deep-Learning Video Cameras
This week at its 4G/5G Summit in Hong Kong, Qualcomm and partner ThunderSoft introduced a reference design for IP video cameras based on an octa-core Snapdragon 625 processor. The kit is intended for products such as home and enterprise security cameras, action cams and 360-degree cameras. The reference design kits have a starting price of $799.
Devices based on the Snapdragon 625 IP Camera reference design can capture 24-megapixel images and 4K video at 30 frames per second: not uncommon for many modern-day cameras. However, this isn’t just about efficient high-definition video capture, but also on-device analytics.
These cameras can have features such as face detection and recognition and object tracking. For example, security cameras could discern between types of objects and use their own on-board reference library to detect individual faces. Although the reference design includes LTE and Wi-Fi connectivity, the concept is to remove reliance on cloud computing power and to move the intelligence into the device.
Using local computing power means reduced latency, particularly in areas with limited connectivity. A camera intelligent enough to understand what needs to be recorded and processed means there is less content to store and transmit, leading to increased efficiency. Inconsequential images don’t need to be pushed out.
Qualcomm’s camera kit currently supports Android, but the company is also adding Linux support to its connected camera line-up. This will provide product developers a broader software ecosystem and more security options.
Qualcomm is calling these devices a new class of “conscious camera” given their ability to monitor and then make decisions about what should be shared and what can be ignored. Deep intelligence is now being built into devices such as security cameras which can act as independent nodes on a network. The Internet of things doesn’t mean a device has to outsource its computing to the cloud. Qualcomm is enabling intelligence in more types of devices, building out the intelligence of things.