Use Case
Optimization of the In-Cabin Monitoring System
Opt-AI has successfully optimized an in-vehicle monitoring system for the target embedded system and RTOS environment proposed by LG Electronics.
We have also secured a mass production optimization project.
Project Goal
-
Optimization of the In-Cabin Monitoring System
Inference Environment
-
HW: ① Qualcomm
② Ambarella HW platform -
OS: QNX RTOS & Linux
Development Details
-
Development Pipeline & Toolchain Construction
-
Model Lightweighting (Quantization)
-
Embedded System Optimization
-
Application Development
Development of Multi-model Inference Scheduling Technology
Opt-AI have successfully developed multi-model inference scheduling technology for the target embedded system and RTOS environment proposed by LG Electronics. We are currently collaborating on productization and further commercialization
Project Goal
-
Development of Multi-Model Inference Scheduling Technology for Processor Resource Optimization
Inference Environment
-
HW: Qualcomm ADP (SA8295)
-
OS: Blackberry QNX 7.1 RTOS
Development Details
-
Model Profiler: Analyzing Model Inference Performance + Searching for Optimal Operators
-
Model Scheduling: Determining the Execution Order and Processor Placement for Multiple Models
-
Task Manager: Providing Inference on Embedded Systems under Scheduling Policy
Development and Optimization of AI Models on Low-End Arm Processor
Through a PoC with Hanwha Systems, Opt-AI have developed and optimized AI models on low-end Arm processors,
exceeding the target performance, and are currently discussing follow-up projects
Project Goal
-
Development and optimization of symbol image extraction and analysis technology in the target embedded system
Development Environment
-
Display Hardware: Azimuth Indicator IP-1310/ALR1
-
Target Processor: i.MX6 4x Arm Cortex-A9, up to 1.2 GHz per core (actual 0.9 GHz)
-
Screen Size: 480 x 480
-
Target Symbol Images: Typical Threat Display Symbol Set
Development Details
-
Building Training Database: Develop a symbol image and video generator
-
Model Development: HW Aware Model Profiling + Design AI Model for Symbol Image Detection + Fine-Tuning
-
Model Compression/Optimization: Quantization (FP16) + SIMD Optimization for Quad-core Processors
