Leveraging the inherent low-power parallelism of the PowerVR Rogue GPU, Imagination Technologies announces a PowerVR imaging framework for Android. This set of software components will allow OEMs to integrate the latest computational photography and computer vision features into their camera applications.
Advanced vision and computational photography features such as HDR (High Dynamic Range), panoramic stitching, gesture recognition and augmented reality all require large amounts of processing power. Today, most OEMs continue to rely on CPU/DSP cores to meet the requisite performance requirements, but these processors struggle to deliver sustained video-rate processing of HD content, largely due to thermal limits of the devices.
Many image processing algorithms are well suited to the massively parallel architecture of the GPU. The PowerVR imaging framework tightly integrates the GPU with other system components such as a camera sensor, ISP (Image Signal Processor), CPU and other SoC-specific hardware to create a programmable image processing pipeline that can be easily incorporated into an OEM’s camera application. With the power and performance efficiency of PowerVR GPUs, Imagination’s OEM partners can deploy cutting-edge vision features into their camera applications, ahead of dedicated hardware implementation in future SoCs, all within the constrained power budget of smartphones, tablets and other consumer devices.
The imaging framework comprises a set of extensions to the OpenCL and EGL APIs that enable efficient zero-copy sampling of YUV and RGB camera data. The extensions enable direct manipulation of YUV data formats for example, which can be used to accelerate algorithms that operate only on luminance data. Another extension allows hardware to be configured to dynamically convert YUV data to RGB when sampled.
The framework also includes low-level functions for integrating these zero-copy extensions within the camera HAL (Hardware Abstraction Layer). Once integrated, these extensions can be used to overcome any limitations of camera lenses, the ISP (Image Signal Processor) hardware feature set, or the stock Android camera application. Customers have already used the zero-copy features in the framework to implement accelerated vision applications, with no need for any additional coherency hardware.
Imagination is working with several software partners to enable them to showcase computational photography and vision solutions today, some of which will be shown at CES. Imagination is uniquely positioned to help its partners build innovative vision systems, with its in-depth knowledge of imaging processors, GPUs, video encoders and embedded CPUs, as well as its expertise in heterogeneous processing and with compute APIs such as OpenCL, RenderScript and OpenGL ES 3.0 and emerging APIs for vision applications such as OpenVX.
“With the imaging framework for PowerVR, we are making it easy for our customers and their OEMs to differentiate their camera applications using their existing SoCs, all without adding to the hardware cost,” commented Peter McGuinness, Director of Multimedia Technology Marketing, ImgTec. “We are making our imaging platform available to lead partners under an early access program. I am excited to see the excellent progress of our software partners in building up the vision ecosystem around this foundational PowerVR computing technology.”
Michael Minkevich, Vice President of Technology Services, Luxoft, added: “Imagination’s imaging framework has allowed us to efficiently implement many computational photography algorithms on the GPU such as denoising, structure and detail enhancement and HDR that help maximise the value for PowerVR users. The reference camera demonstration app included as part of the imaging framework has allowed us to demonstrate near-productised solutions to end OEMs.”