使用骁龙计算机视觉技术开发下一代相机应用.pdf

1、Develop Next-Gen Camera Apps Using Snapdragon Computer Vision TechnologiesJudd HeapeVP,Product ManagementCamera,Computer Vision and VideoEVA:Engine for Visual Analytics Video bokeh Background replacement Cinematic video XR:depth and 3D reconstruction Face beautification Face-centric 3A Cinematic vid

2、eo AI segmentation camera Scaling,rotation and warping Lens distortion correction Rectification 3D zoom SLAM and tracking VR obstacle tracking Autonomous parkingIndirect Time of Flight sensor(iToF)Face detectionFace landmark detectionFeature extraction Accurate depth Face authentication Auto FocusDe

3、pth processingEngine for Visual Analytics Frame alignment Multi-frame noise reduction Motion-compensated temporal filteringMotion estimation2 2023 Qualcomm Technologies,Inc.Geometric correctionEVA architectureEngine for Visual AnalyticsComputer Vision HW acceleration Power savingsPerformance enhance

4、ment Access EVA APIs from CPU or Hexagon Tensor Processor(HTP)Offload NPU/DSP/GPU/CPU for CV workloadsEVA driverCV algorithmsEVA driverCV algorithmsCV applicationsEVA-CV HW acceleratorHTPReprojection and Grid InversionDepth estimationGeometric correctionMotion estimationFeature extractionVideo Proce

5、ssing UnitGPUNPUFastCVCamera ISPCPUiToFprocessingDDR-SDRAMXR acceleration3 2023 Qualcomm Technologies,Inc.EVA access EVA blocks function from memory-to-memory for maximum flexibilityAPIs(for both CPU and HTP)include both synchronous APIs and asynchronous APIsThere are direct interrupts between HTP a

6、nd EVA cores for low-latency communicationEVA includes an embedded processor primarily for task scheduling and managing hardware pipesEVA hardware pipes are shared between certain functions EVAISP front endISP processing engineDDR-SDRAMEVAISPExternal appDDR-SDRAMEVAExternal sourceExternal appDDR-SDR