Application overview
Front camera systems are essential for modern Advanced Driver Assistance Systems (ADAS), enabling critical autonomous driving features. These vision-based systems must detect and classify multiple objects simultaneously and in real-time under diverse driving conditions.
Multi-object detection in automotive environments presents unique challenges. Objects often vary significantly in size and distance, lighting conditions can change rapidly and safety-critical decisions require low-latency processing. Neural network-based detection provides the robustness needed to handle these challenges, but requires substantial computational power to maintain the level of performance necessary for safe vehicle operation.
TI’s C7™ NPU equipped processors in the TDA4x family deliver the compute performance needed for real-time front camera processing while meeting automotive safety and reliability standards. The integrated approach enables ADAS features like adaptive cruise control, automatic emergency braking, lane keeping assist and traffic sign recognition.
Starting evaluation
Data collection
Front camera ADAS systems require extensive training data captured under diverse real-world conditions covering varied scenarios such as, urban, highway and rural roads; day, night, dawn and dusk lighting; clear, rain, fog and snow weather conditions; different traffic densities; and diverse geographic locations. Data should be collected from cameras matching the production specifications, including resolution, field of view, lens characteristics and mounting position.
Training data requires bounding box annotations identifying objects with class labels. For lane detection and road boundary segmentation, pixel-level annotations may be required.
Data quality assessment
Image quality directly impacts detection performance and therefore overall system safety. Assess that the camera captures sufficient detail for small or distant objects. Special attention should be paid to safety-critical objects like pedestrians and cyclists. Missing annotations or incorrect classifications in training data will propagate to model performance, potentially creating safety hazards.
Build and train your model
TI provides pre-trained models optimized for TDA4x and AM62x devices in the Model Zoo, ready for evaluation and deployment. These models are trained on large-scale datasets and support common ADAS tasks including multi-object detection and semantic segmentation.
For custom model development, developers can use Edge AI Studio or edgeai-modelmaker for training TI-supported neural network architectures. More advanced developers may use frameworks like PyTorch or TensorFlow to develop custom models, then export to ONNX or TFLite formats for deployment on TI processors.
Find the right model for your needs
Model selection depends on the specific ADAS features required and performance constraints. For systems requiring simultaneous detection of multiple object types, multi-class object detection models like YOLOX, SSD-MobileNet or EfficientDet provide good accuracy-latency tradeoffs.
Benchmark data in Edge AI Studio helps evaluate model performance on target hardware. Key metrics include inference latency, FPS throughput and detection accuracy.
Deploying your model
Model deployment requires compiling the trained model for TIDL acceleration on C7x+MMA cores. Models in ONNX or TFLite format are compiled using edgeai-tidl-tools on the TI GitHub. The compilation process optimizes layer execution, quantizes weights and configures multi-core splitting for maximum throughput.
Model artifacts are deployed through runtimes like ONNX Runtime, LiteRT (formerly tensorflow-lite) and TVM using TI Deep Learning (TIDL) as the backend software for hardware acceleration.
The complete front camera pipeline integrates multiple processing stages including: camera capture, VPAC pre-processing, TIDL AI inference and CPU post-processing.
For application development, TI provides either GStreamer or OpenVX based frameworks for composing the complete pipeline with multiple hardware accelerators. Demo applications developed by TI partners like Stradvision and PhantomAI demonstrate integration with ADAS control algorithms and visualization for development and validation.
Choosing the right device for you
Device selection for front camera ADAS systems depends on the number and type of AI tasks, camera resolution, required frame rate and functional safety requirements. The TDA4x family of automotive grade processors provides multiple options optimized for vision-based ADAS applications. Refer to the table below for performance comparison across different devices (benchmarks were produced using SDK version 10.1). Note: For comprehensive benchmarks of these devices, use the model selection tool available on Edge AI Studio.
| Product number | Processing core | NPU performance |
|---|---|---|
| TDA4VH-Q1 | 8x Arm® Cortex®-A72 + 4x C7™ NPU | 32 TOPS |
All the hardware, software and resources you’ll need to get started
Hardware
AM62A
The AM62A is the lowest-cost AI-accelerated device in the AM6xA family, and is well suited for evaluation. A generic USB camera or webcam can be used for image capture and model evaluation on live data.
TDA4AEN
The TDA4AEN is the entry level device in the TDA4x family. The 4 TOPS of AI performance enable running object detection models for front camera ADAS efficiently.
Software & development tools
The Edge AI processor SDK is Linux-based and includes the necessary software components to run a compiled model with hardware acceleration. Other Edge AI accelerated processors may be substituted for AM62A.
This tool contains tools for training, compiling and deploying a model to TI Edge AI processors. A model selection tool is available to view pre-generated benchmarks of popular models.
Tools for micro processor devices with Linux and TIDL support. TI's edge AI solution simplifies the whole product life cycle of DNN development and deployment by providing a rich set of tools and optimized libraries.
Supporting resources
The Vision Apps framework included in the Processor SDK provides a front camera demo application as a reference implementation. This user guide provides a brief overview and instructions for running the front camera demo.
Object detection in vision systems with 25x lower latency
Find and localize specific objects, people in real-time at high framerate with AI-accelerated processors and industry-standard software
In-Cabin Safety: Edge AI for Driver Monitoring Systems (DMS)
Enable Euro NCAP compliant DMS using real-time multi-task vision AI to detect drowsiness, distraction & head pose at the edge.
Bird's Eye View (BEV) in L2/L2+ advanced driver assistance systems (ADAS)
High-performance whole-vehicle perception with BEV networks on TI's AI-accelerated processors