Professional Documents
Culture Documents
Slyp 771
Slyp 771
解決方案設計簡介
TI 資深汽車處理器應用工程師
Rio Chan
1
Designing Deep Learning Solutions
for ADAS
2
Outline
• What is ADAS?
• What are the important peripherals for ADAS?
• What Deep Learning can do for ADAS?
• TI TDA4 ADAS design concept?
• TI TDA4 road maps?
• TI Deep Learning.
• TIDL Demos?
• What TI can support?
3
ADAS?
4
ADAS level
5
WHAT ARE THE IMPORTANT
PERIPHERALS FOR ADAS?
6
What are the important peripherals for ADAS?
TI TDA4
https://www.ti.com/product/TDA4VM
7
What are the important peripherals for ADAS?
TI Radars
https://www.ti.com/sensors/mmwave-
radar/overview.html
8
What are the important peripherals for ADAS?
9
What are the important peripherals for ADAS?
11
TI Processor SDK | Single development kit for all families
TI-RTOS
Sensor
Cockpit Cluster Amplifier Powertrain Gateway Camera Radar LIDAR fusion
AM2x AM6x DRA8x TDA4x
Jacinto
DRA 13
Jacinto’s TDA4 benefit for “ADAS” system
Jacinto
TDA 14
TI Jacinto™ 7 | Heterogeneous compute platform Unique TI
innovation
4K
L3$
GPU
Optimize entire platform around ASIL-D Memory 64b ARMv8
programmer productivity on the Safety MCU System MPUs
More Processing
MPUs
Performance
Dual-core Dual-core
•
•
Jacinto™ automotive
Common high-performance chassis
High-bandwidth coherent SoC interconnect for main processor & memory
Packet-based interconnect for latency-sensitive interfaces (Ethernet, CSI)
A72 cluster A72 cluster
•
processors
Dedicated boot engine (early boot)
Automotive connectivity 15
• CAN-FD, MOST-150, FlexRay, Ethernet AVB, USB 3.1 Dual-core Dual-core
• GbE switch, PCIe Gen4 with switch RF5 cluster RF5 cluster
Embedded Machine Learning Architectures
16
17
TDA4 ADAS Design concept
Surround view / Parking
Front Camera
TDA4
ADAS Domain controller
Radar
18
19
20
21
22
23
24
TDA4 ROADMAP
25
Jacinto Vision Analytics Portfolio
26
TI DEEP LEARNING
27
What Deep Learning can do for ADAS?
28
Edge inference software stack on TI SoC
Model artifacts
TIDL Optimizer: Layer fusion | Post Training Quantization | Range Calibration | Memory optimizations
C7x + MMA
Cortex-A
(Deep Learning Accelerator)
29
User Application Jacinto 7 processor
Python / C / C++
TFLite/ONNX-RT/Neo-AI-DLR
API | interpreter | scheduler
+ - * =
Linux OS
C7x DSP with MMA*
Open-source
ARM ARM
Cortex
Cortex
A72 A72
CPU kernels Deep learning accelerator
30
31
TI Deep Learning | SW Eco System
TIDL Importer
➢Enables development on Open frameworks
Training frameworks Model Formats ➢Model Parameters Quantization & Calibration TIDL Inference
➢Layers Merging/Fusion ➢Abstracts embedded development
TIDL Graph Compiler ➢High Efficiency Inference Implementation
➢Memory management, Network Execution
Plan/order OpenVX Node
➢Models the device & Estimates performance
Frozen-pb/ ➢Deep learning Use case development
TIDL Host Emulation
TFLite ➢PC/Host emulation of Inference for offline
across cores (IPC, Pipelining )
Validation and Accuracy benchmarking
TIDL Inference
TIDL Host Emulation
Prototxt
Caffemodel
Training Inference
32
What Deep Learning can do for ADAS?
33
Real-time 8x 2MP camera input demo on single ISP
• TDA4VM Camera performance as below.
34
DMPAC: Depth and Motion Perception Accelerator
Stereo disparity map 90 Mpix/s, <0.5W Dense optical flow 150 Mpix/s, <0.5W
Up to 2MPix Upto 2MPix
Stereo Dense
Depth Optical
Engine Flow
Stereo Depth Engine: for each location <x,y> , estimates depth by Dense Optical Flow: for each location <x,y>, estimates 2D
measuring disparity from 2 different views motion vector field given two images
Assign confidence score for each disparity output Generate confidence score for each flow vector output
35
4x 1MP @ 20 fps
disparity map per core
TDA4 EVM AND DEMOS
36
TDA4V Mid EVM Details
Baseline Hardware Kit Modular Development System
➢Provides complete access to Jacinto™ 7 SoC
➢ Processor SOM ➢HW platform for SDK Validation
➢New SOM for each Jacinto™ 7 SoC Variant
➢On-board XDS110 JTAG for debug via CCS
➢XDS560v2 for advanced debug and trace
➢ Common http://www.ti.com/tool/TMDSEMU560V2STM-U
Baseboard
Optional Expansion cards for application needs
37
Jacinto™ 7 TDA4VM analytics and vision demos
Front camera analytics with TDA4 Smarter Parking with TDA4 TDA4: Multi-Camera vision processing
Real-time 8-Camera
Input to TDA4
Deep Learning
Surround View
Based Forward
with Analytics
Camera Analytics
➢ High-quality 3D Animated Car Model, overlays
➢ Semantic Segmentation and Free space ID ➢ Eight 2MP cameras feeding single TDA4VM.
for projected path and distance warning.
➢ Multi-class Object Detection ➢ All camera images pre-processed by single
➢ Transparent car view using history
➢ Custom DL networks built on TI Deep Learning ISP in real-time
➢ Object Detection, Lane ID, Parking spot ID
Libraries (TIDL) and running on C7x+MMA ➢ All code freely available in TI’s Processor SDK
using DSP based algorithms.
➢ AVP with Semantic segmentation, object and ➢ Ruggedized Development platform for
➢ NCAP ready multi-class object detection using
parking spot detection based on TIDL applications on TDA4VM SoC.
TIDL Libraries and C7x+MMA
➢ Simultaneous Real-time 4x 2MP camera ➢ Multi-sensor interfaces: camera, radar, lidar…
➢ Fusion with IMU and GPS for Localization
Surround view with 3D car model. ➢ Software compatible with TI EVM based on
➢ Concurrent applications on one TDA4VM SoC
➢ All code freely available in TI’s Processor SDK Processor SDK 38
Jacinto™ 7 TDA4VM YouTube Videos
Front camera analytics with TDA4 Smarter Parking with TDA4 TDA4: Multi-Camera demo
Real-time 8-Camera
Input to TDA4
Deep Learning
Surround View
Based Forward
with Analytics
Camera Analytics
➢ https://youtu.be/F4KNWuTSJds
➢ https://youtu.be/F4KNWuTSJds ➢ https://youtu.be/zsYbPPhO0t8
➢ https://youtu.be/PER77CgoHWk ➢ https://youtu.be/bOSmeej9_cQ
➢ https://www.businesswire.com/news/home/202
00107006173/en/
39
WHAT TI CAN SUPPORT?
40
Software and tools - overview
41
TDA4 AutoSAR
42
TDA4 PinMux tool
https://e2e.ti.com/support/processors/f/processors-forum/927526/faq-ccs-tda4vm-pinmux-guide-for-jacinto-processors
43
TDA4 Cloud AI Tool: https://dev.ti.com/edgeai/
44
TDA4 Cloud AI Tool: https://dev.ti.com/edgeai/
45
TDA4 Cloud AI Tool: https://dev.ti.com/edgeai/
46
TI ADAS system of experienced partners
OPERATING SYSTEMS AND SECURITY SAFETY SOFTWARE, CERTIFICATION
MCAL
AutoSAR Classic
Adaptive AutoSAR
E NGINEERING S ERVICES
48
IMPORTANT NOTICE AND DISCLAIMER
TI PROVIDES TECHNICAL AND RELIABILITY DATA (INCLUDING DATASHEETS), DESIGN RESOURCES (INCLUDING REFERENCE
DESIGNS), APPLICATION OR OTHER DESIGN ADVICE, WEB TOOLS, SAFETY INFORMATION, AND OTHER RESOURCES “AS IS”
AND WITH ALL FAULTS, AND DISCLAIMS ALL WARRANTIES, EXPRESS AND IMPLIED, INCLUDING WITHOUT LIMITATION ANY
IMPLIED WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE OR NON-INFRINGEMENT OF THIRD
PARTY INTELLECTUAL PROPERTY RIGHTS.
These resources are intended for skilled developers designing with TI products. You are solely responsible for (1) selecting the appropriate
TI products for your application, (2) designing, validating and testing your application, and (3) ensuring your application meets applicable
standards, and any other safety, security, or other requirements. These resources are subject to change without notice. TI grants you
permission to use these resources only for development of an application that uses the TI products described in the resource. Other
reproduction and display of these resources is prohibited. No license is granted to any other TI intellectual property right or to any third party
intellectual property right. TI disclaims responsibility for, and you will fully indemnify TI and its representatives against, any claims, damages,
costs, losses, and liabilities arising out of your use of these resources.
TI’s products are provided subject to TI’s Terms of Sale (https:www.ti.com/legal/termsofsale.html) or other applicable terms available either
on ti.com or provided in conjunction with such TI products. TI’s provision of these resources does not expand or otherwise alter TI’s
applicable warranties or warranty disclaimers for TI products.IMPORTANT NOTICE
Mailing Address: Texas Instruments, Post Office Box 655303, Dallas, Texas 75265
Copyright © 2021, Texas Instruments Incorporated