Design of Advanced Driver Assistance System Based on TMS320DM6437 Digital Media Processor

Cars have become an indispensable part of modern society, providing a low-cost means of transportation for the transportation of goods and people around the world. Unfortunately, a large number of accidents and casualties occur every day due to vehicle driving. According to statistics from the World Health Organization, the number of deaths in traffic accidents worldwide reached 1.2 million in 1998; in 1990, traffic accidents were the ninth leading cause of death in the world, and it is expected that by 2020 it will further rise to the third leading cause of death.

For these reasons, automotive original equipment manufacturers (OEMs) and their supplier partners, as well as global government agencies, have been working hard to develop and promote active safety and advanced driver assistance systems (ADAS), which are designed to reduce the incidence of accidents and reduce the severity of collisions. Automotive industry analysts believe that by 2010, ADAS will become the top new technology, which can not only help drivers realize potential dangers in advance, but also potentially extend the driver's reaction time through technologies such as lane departure warning systems (LDWS), drowsiness detection and night vision systems.

As consumers gain a deeper understanding of ADAS and learn that it can provide greater safety, it is expected that they will be more receptive to the technology, thereby driving the growth of market demand for this technology. ADAS has been applied to luxury cars, and as the technology matures, it will gradually enter the mass market and be applied to ordinary vehicles. At the same time, the increase in production will significantly reduce product costs. For automotive OEMs, given that passive safety systems are rapidly becoming standard technology for automobiles, active safety and ADAS technology will also help to achieve distinctive and specialized products.

ADAS Overview

ADAS is not designed to control the vehicle, but to provide the driver with relevant information such as the vehicle's surroundings and vehicle operating conditions, reminding the driver of potential dangers, thereby improving driving safety.

ADAS applications use a variety of sensors to collect physical data about the vehicle and its surroundings. Once the data is collected, the ADAS system uses processing techniques such as object detection, recognition, and tracking to assess the risk. Two application examples are lane departure warning systems (which detect unintentional lane departures and immediately alert the driver) and traffic sign recognition. When lane departure warning is enabled, the system detects and tracks lane conditions based on the vehicle's position and notifies the driver if the vehicle crosses the line into an adjacent lane. In the case of traffic sign recognition, the system can identify traffic signs and tell the driver the current maximum speed limit or notify the driver of the specific section of the current driving.

Different systems often require different types of sensors to collect environmental information. For example, lane departure warning systems use CMOS camera sensors, night vision systems use infrared sensors, adaptive cruise control systems (ACC) usually use radar technology, and parking assistance systems use ultrasonic technology. Although the technical details of different applications vary, the technical process usually includes three stages: data acquisition, preprocessing, and postprocessing. The preprocessing stage performs full image processing functions, which is data-intensive and has a more conventional structure. Specifically, it includes image transformation, stability, feature signal enhancement, noise reduction, color conversion, motion analysis, etc. The postprocessing stage performs feature tracking, scene interpretation, system control, and decision making (see Figure 1). Regardless of the type of sensor used to collect environmental information, a data set that generates a basic image can be obtained.

Figure 1: The entire process of active safety system

Identifying, tracking, and evaluating driving-related objects is a complex task. Driving style and conditions affect the quality of the raw data collected by sensors and can obscure important details needed to identify and track objects. Drivers are highly dynamic and unpredictable in different weather conditions (strong sunlight, rain, fog, snow, etc.). When faced with more complex situations, all data must be processed in real time with a processing delay of no more than 30ms. A warning delay of half a second is likely to cause an accident without the driver being able to react to the warning in time.

Each step from data acquisition to action requires powerful signal processing capabilities, so timely and accurate execution of active safety and ADAS systems must require high-performance products to support them. Digital signal processors (DSPs) designed and optimized for automotive safety applications, such as the TMS320DM643x DaVinci processors from Texas Instruments (TI), can provide the required performance, enabling OEMs to bring active safety and ADAS technologies to market.

Dynamic flexibility

In addition to high performance, ADAS applications also require a flexible architecture to meet a variety of functional requirements. For example, traffic signs in different countries vary from language, text fonts, shapes to colors. We must provide enough flexibility so that we can reuse the technology as much as possible between different product lines and promote technological innovation at a normal speed and low cost according to the requirements of emerging markets. Software innovation is the most efficient way. The Da Vinci processor adopts a software-programmable architecture, so it can provide developers with enough flexibility to support changing algorithms.

Let's imagine the pre-processing algorithm without taking into account driving conditions (such as strong sunlight, etc.). Some automotive suppliers use one algorithm, while others use one algorithm for daytime driving and another algorithm for nighttime driving. In fact, we will need a variety of different pre-processing and post-processing algorithms for many different driving conditions. The system must also be able to adapt quickly, such as switching from daytime driving mode to nighttime driving mode when the vehicle enters a tunnel.

Please note that different sensors in a vehicle should perform different functions (see Figure 2). For example, the side sensors detect blind spots; the forward sensors are responsible for vehicle, lane, traffic sign and pedestrian recognition; the in-car sensors detect whether there is someone in the car, whether the driver is drowsy and what he wants to do, etc.

Figure 2: A variety of sensors installed in a vehicle perform specific tasks

In addition, different sensors should process different types of data. Some traffic sign recognition algorithms perform mainly on the color of the sign, in which case the forward sensor should support a wide color range. On the other hand, grayscale sensors are more sensitive to changes in brightness and have almost twice the spatial resolution of color sensors. The implementation of most ADAS functions mainly depends on the sensitivity of the sensor, so grayscale cameras are more suitable. It is also important to note that image sensors for ADAS applications usually have a higher dynamic range, generally more than 8 bits per pixel.

The most effective way to solve the technical problems is to execute multiple algorithms on a single DSP. For example, a forward image sensor can provide the video information required for lane departure warning and traffic sign recognition at the same time. Ideally, a single DSP can perform pre-processing and multiple recognition tasks under all driving conditions, such as lane departure warning and traffic sign recognition, which helps to streamline the number of chips, thereby reducing failure points, improving system reliability, and reducing system costs, which are all critical to the development of automotive applications.

To improve the robustness of ADAS, we also need to coordinate between all active safety subsystems within the vehicle. For example, the direction and focus of the driver's attention will directly affect the effectiveness of the traffic sign warning. For example, the driver should be able to notice the sign ahead and slow down under normal driving, but the system issues a warning too early, which is not only not conducive to driving, but also creates trouble for the driver. Therefore, before issuing a warning signal, we need to collect information from the traffic sign recognition system and the in-vehicle driver monitoring system at the same time. If the driver monitoring system reflects that the driver is facing the road, then we don't have to issue a warning about the stop sign ahead immediately.

A robust ADAS system can even assess complex driving situations. For example, if the car is approaching a stopped or slowing vehicle quickly, an emergency lane change may be necessary. In this case, the lane-offset warning should be suspended to prevent the driver from being distracted during the lane change. Of course, if the blind spot monitoring system detects that there are other vehicles next to the car, the system should issue a warning.

System-on-chip architecture improves design efficiency

Current system-on-chip (SoC) architectures integrate all the peripherals required for the entire video/image processing system on a single chip, further improving design efficiency. Due to the support of a variety of peripherals, today's highly integrated devices can also be easily connected to other parts of the vehicle system. For example, the SoC can provide direct video output for applications such as parking assist rearview cameras, or it can be directly connected to the vehicle's main control system via appropriate buses such as CAN, LIN or FlexRay.