Hailo-15 is a family of AI Vision processors for smart cameras that deliver up to 20 TOPS of AI inference and are able to process deep-learning AI applications such as video analytics.
The Hailo-15 appears to build upon the earlier Hailo-8 AI processor with up to 26 TOPS and more limited computer vision capability. The new AI vision processor integrates a quad-core Cortex-A53 CPU sub-system, can handle 4K video streams, and features “advanced computer vision engines”. So the main differences are that the Hailo-15 is a complete SoC / standalone processor and is specifically designed for AI cameras.
Hailo-15 highlights:
- CPU – Quad-core Arm Cortex-A53 processor up to 1.3 GHz or 1.1 GHz (Hailo-15L)
- DSP – Vector DSP, 256 MACs @ 700 MHz supports up to 350 GOPs
- AI accelerator / NN core sub-system with dataflow architecture
- Hailo-15H – Up to 20 TOPS
- Hailo-15M – Up to 11 TOPS
- Hailo-15L – Up to 7 TOPS
- ISP – Up to 12MP resolution, 600 Mpixel/s pixel rate with High Dynamic Range (HDR) and Noise Reduction (NR) algorithms
- VPU – Video Encoding (HEVC & AVC) H.265/H.264, multiple streams
- Memory I/F – LPDDR4/4X 32bit @ 4266 MT/s or 3200 MT/s (Hailo-15L)
- Storage – eMMC 5.1, SDIO 3.0 up to HS200, and QSPI
- Video Output – MIPI DSI
- Video Input – dual MIPI SI, 24-bit DVP
- Networking – Gigabit Ethernet with RMII/RGMII
- USB – USB 3.1 Gen2 Host/Device, USB 2.0 host
- Other Peripheral/Sensor interfaces
- PCIe Gen 3.0 x4 (Endpoint or RC)
- 4x I2C, 4x SOU
- 2x I2S in, 2x I2S out
- Security – Secure boot and debug with hardware-accelerated crypto library, TrustZone, TRNG, and Firewall
- Power consumption – “Standard” camera power consumption (and cost envelope) suitable for fanless camera
- Package – 15x15mm FCBGA
- Temperature Range – -40 to +85°C
Hailo says the 20 TOPS Hailo-15H is capable of running the YoloV5M6 object detection model with 1280×1280 input resolution in real-time, or the ResNet-50 classification model at 700 FPS.
The AI vision processor will be integrated into Smart cameras used in Smart Cites to quickly detect and respond to incidents, in factories to increase productivity and machine uptime, by retailers to protect supply chains and improve customer satisfaction, as well as transportation authorities to locate lost children or luggage and detect accidents.
The company also offers a camera development kit with an Hailo-15 single board computer with dual-camera, HDMI, Ethernet, USB ports, audio jacks, and expansion header, a Yocto-based Linux distribution with a full stack DSP, libraries (OpenCV, GStreamer…), BSP, and drivers, as well as documentation and support.
The Hailo-15 does not seem to be available just yet based on the rendering of the devkit, but Hailo will be showcasing its new AI vision processor at ISC-West in Las Vegas, Nevada, from March 28-31, at booth #16099. More details can be found in the press release and the product page where you can also order the development kit.
Thanks to TLS for the tip.
Jean-Luc started CNX Software in 2010 as a part-time endeavor, before quitting his job as a software engineering manager, and starting to write daily news, and reviews full time later in 2011.
Support CNX Software! Donate via cryptocurrencies, become a Patron on Patreon, or purchase goods on Amazon or Aliexpress
