ADAS Solution for Electric Vehicles

A real-time data processing system for 15 Full HD in-vehicle cameras

  • Texas Instruments
  • Linux
  • RTOS
  • C/C++
  • DSP
  • GMSL
  • FDP-Link
  • Linux SMP
Solution ADAS platform for camera data analysis
Industry Automotive
Engagement model T&M (time and materials)
Methodology Agile
Team
  • Embedded Engineers
1
2
3
4
5
6

Customer

Case Highlights

  • Designed to processes data captured from 15 Full HD cameras in real time
  • Linux SMP architecture
  • 3 x FPD-Link III displays
  • Five ARM-powered processors
  • Handles exposure, autofocus, white balance, and fish-eye effect

Problem

The customer specializes in the production of lightweight commercial vehicles. They wanted to enhance their products with an advanced driver assistance system (ADAS). The solution will analyze data captured by 15 vehicle cameras to avoid collision and monitor weather conditions.

The client tasked Softeq with designing a platform that could be used to implement any business logic — from notifying the driver about the situation on the road to autonomous driving.

Solution

Market Research

Softeq researched the CPU modules available on the market.

First, we were looking for a system on a module (SoM) device, which would serve as a basis for our system. The device had to meet the following requirements:

  • Be capable of processing data captured from 15 Full HD cameras in real time
  • Collect data not only from cameras but from other internal vehicle systems (HMI, IVI, ECU and more) via CAN, LIN, etc.

We evaluated 10+ CPUs from different vendors. The results showed that there wasn’t an SoM solution on the market that could handle more than four Full HD cameras in real time, including post-processing, so we decided to build one from scratch.

Solution Design

Softeq designed a custom solution based on the Linux symmetric multiprocessing (SMP) architecture, which is able to combine multiple processors into a single system.

The solution consists of five ARM-powered processors: four for processing digital images generated by the onboard cameras, and one for mathematical calculations and external connection processing. To take the load off the main processors, the device incorporates coprocessors responsible for image analysis. 

The system can use RADAR/LiDAR technology to precisely estimate the distance between objects. The design includes 15 camera inputs and 3 FPD-Link outputs for displaying data. It supports Automotive Ethernet, CAN, LIN, and more.

Data Transfer

Softeq developed drivers for FPD-Link III interfaces.

As the automotive industry requires specialized technical solutions for data transfer - data buses, we used FPD-Link III interfaces for input and output devices (cameras and displays). Our team developed custom drivers that allow the system to interact with light sensors. At this level, lens correction algorithms are applied to enhance images (exposure, auto focus, white balance, cropping, scaling, fish-eye effect reduction, and color space conversion).

Device for Environment Monitoring

In order to make sure the idea was viable, Softeq built a monitoring device for collecting information about the surroundings.

The CPU-based device connects to several rear-view and parking cameras and collects the data about the environment. The solution performs basic image signal processing of data captured by camera light sensors. For this, it uses our lens correction algorithms.

Results

Viable system design

Softeq designed a solution that would process data captured by 15 Full HD cameras in real time. Currently, the team is working on improving the lens corrections algorithms.