Every 13 seconds. That’s how often a burglary or break-in occurs in the U.S.
The migration of security cameras from analog CCTV to IP-based digital cameras has opened up newer possibilities to an enhanced user experience through advanced connectivity and smarter computing. Big strides in edge computing and on-device processing are expanding the role of cameras in securing large buildings and industrial installations. For consumers, that same IP camera technology is leading to affordable access to home security systems that were unavailable just a few years ago.
Sending wisdom, not just data
Those IP cameras have improved and evolved with security systems through several steps:
- Basic sensors — The first generation of IP cameras used for security acted as basic sensors, sending digital streams back to a storage device or cloud for manual analysis. Detecting a one-minute intrusion or break-in could require searching through hours of footage.
- Cloud-based AI — The introduction of cloud AI took the tedious task of watching hours of video and enabled smarter video analytics. But privacy concerns, network latency and the cost of cloud computing remained sticking points.
- Smart network video recorders — Along came NVRs and smart NVRs. They moved the inferencing step of machine learning from the cloud to an edge server, addressing privacy concerns and cloud costs. However, latency remained a problem.
- Intelligent edge — The convergence of technologies like AI, advanced image signal processing (ISP), connectivity and security has helped bring the “smarts” to the edge. Cameras can now see in low light, analyze video without going to the cloud, make smart decisions and send alerts in real time through a secure, low-latency communication link. Powerful yet power-efficient mobile platforms, high-resolution video and advanced connectivity are the main drivers for these capabilities.
Main challenges to solve
- Useful inferencing requires high-resolution video, even in low ambient light and highly dynamic conditions.
- Effective video analytics can support use cases like detecting firearms and suspicious objects, blacklisting/whitelisting, generating heat maps and modeling crowd flow — to infer and generate meaningful wisdom.
- Ubiquitous connectivity is essential for transmitting inferences and triggering required actions.
- Security is paramount. It is a security nightmare to have cameras hacked, with the prospect of an intruder gaining access to a live feed.
Our approach, based on the Qualcomm® Vision intelligence Platform, consists of 3 phases:
Phase 1: Perceive
The Qualcomm Vision Intelligence Platform has highly programmable dual ISPs (up to 16MP + 16MP sensor support) and can capture, play back and stream premium 4K video at up to 60 frames per second (fps) with high-efficiency video coding (HEVC). The platform supports staggered high dynamic range (sHDR), electronic image stabilization (EIS), low-light noise reduction, lens distortion correction (LDC) and chromatic aberration correction (CAC).
The Qualcomm® Connected Camera SDK provides APIs for designing advanced camera applications and optimizing access to the rich ISP. An ecosystem of ISVs and ODMs is ready to support and tune MIPI-enabled sensors.
Phase 2: Analyze
The Qualcomm® Snapdragon™ Mobile Platform powers AI through its heterogenous compute architecture: CPU, Qualcomm® Adreno™ GPU and Qualcomm® Hexagon™ DSP with dual Hexagon Vector Extensions (HVX). The combination is designed to provide up to 2.1 teraOPS with low power consumption. The Qualcomm® Neural Processing SDK for AI offers the flexibility to port deep neural networks (DNN) with support for Caffé/Caffe2 and TensorFlow. An ecosystem of AI video analytics companies can implement end-to-end security use cases, including face detection/recognition, object detection and license plate recognition.
Phase 3: Act
The Qualcomm Vision Intelligence Platform provides multiple integrated and system-on-module (SOM)-based connectivity options, including cellular (4G/5G), 2x2 802.11ac Wi-Fi, Bluetooth 5.0, integrated Global Navigation Satellite System (GNSS), audio and RGMII for reliable, low-cost communication. Security features include a trusted execution environment, hardware crypto engines, wireless protocol security and secure boot from a hardware root of trust.
Start building our technology into your smart security inventions
Smart security systems depend on AI, high-resolution video, network connections and powerful yet power-efficient mobile processors.
Ready to start building your own security solution? We have multiple reference designs and development kits available through our ecosystem that provide a quick way to start developing AI video analytics or designing smart security cameras.
Here are a few ways to get started:
- Altek’s Qualcomm®® QCS605-based 360 camera reference design is designed to allow you a 360-degree view of the surroundings by stitching streams from dual IMX577 sensors on the QCS605 edge. The camera can run a DNN on the stitched video in real time for use cases like counting people.
- Altek’s Qualcomm®® QCS603-based bullet camera reference design includes a single IMX334 sensor. Its support for power over Ethernet (PoE) and 1x1 Wi-Fi provides a reference suitable for an indoor enterprise IP camera.
- The Vision AI Developer Kit is a security camera reference design that works as a Microsoft Azure IoT Edge device for scalability through retraining new models on the Microsoft cloud.
- Multiple open development boards are available from Thundercomm and Intrinsyc.
- The Qualcomm Neural Processing SDK for AI is made for model conversion and execution to build video analytics for security.
Looking for inspiration? The Cisco Meraki MV family of cloud-managed smart cameras is based on the Qualcomm Vision Intelligence 100/200 Platform. They can eliminate the need for NVRs and offer advanced DNN video analytics on the edge, sending only metadata to the cloud. The Bosch Open Safety and Security Alliance is enabling an ecosystem of developers, OEMs, ODMs and SoC vendors to define a framework of standards and specifications for common components and drive improved levels of performance for security systems. The Qualcomm Vision Intelligence 300/400 Platform is an Android-based platform to run advanced applications for a differentiated enterprise security solution with shorter time to market.
Multiple OEMs, ODMs and ecosystem players are building our technology into their inventions and you can, too. Learn more about the Qualcomm Vision Intelligence 100/200 Platform and the Qualcomm Vision Intelligence 300/400 Platform.
And keep an eye on this blog. You’ll soon see more news about use cases for our camera technology, in applications like smart fleet management, retail and industrial.
Qualcomm Vision Intelligence Platform, Qualcomm Snapdragon, Qualcomm APQ8053, Qualcomm Neural Processing SDK, Qualcomm Adreno, Qualcomm Hexagon, Qualcomm Connected Camera, and Qualcomm QCS603, and are products of Qualcomm Technologies, Inc. and/or its subsidiaries.