From the earliest cruise control to today's comprehensive safety system covering multiple scenarios such as urban congestion, intersections, and cyclist protection, active safety-assisted driving is actually a continuous improvement and upgrade of the "perception-decision-execution" closed loop.
Fundamentals of Active Safety Assisted Driving Technology
1) Environmental perception
To achieve accurate warnings and interventions, active safety-assisted driving requires a sufficiently comprehensive and reliable perception of the vehicle's surroundings. In this process, millimeter-wave radar, cameras, ultrasonic sensors, and lidar each perform their specific functions while complementing each other.
Millimeter-wave radar emits continuous waves in the 24GHz or 77GHz band, utilizing the Doppler effect to quickly measure the distance and speed of moving targets in front and behind, maintaining stable output even in adverse weather conditions such as smoke, rain, and snow. Cameras, on the other hand, capture rich image details at high resolution, using specially optimized convolutional neural networks to accurately identify pedestrians, vehicles, traffic signs, etc., although HDR fusion and noise reduction algorithms are needed to improve quality in nighttime or backlit scenes. Ultrasonic sensors, while only operating at extremely close range (approximately 0.2–5 meters), have become an indispensable supplement in parking and low-speed environments due to their low cost and high reliability. In recent years, lidar, with its 360-degree three-dimensional point cloud scanning capability, combined with inertial measurement units for spatiotemporal synchronization, can generate three-dimensional environmental models with centimeter-level accuracy, providing the most intuitive spatial information for complex scenes. The signals from each sensor must undergo RF front-end or image preprocessing, analog-to-digital conversion, filtering, and feature extraction before being input into subsequent intelligent algorithms.
2) Multimodal fusion
Based on the parallel operation of multiple sensors, one of the core technologies of active safety-assisted driving is how to fuse this data and extract the most reliable environmental picture. This process usually divides the fusion into several layers: at the bottom layer, the radar range-velocity matrix and the lidar point cloud data are aligned to the same coordinate system to deepen the spatial perception of obstacles; in the middle layer, each pedestrian or vehicle is labeled with higher confidence by matching the target identified by the camera and the trajectory tracked by the radar; at the top layer, the risk assessment network incorporates the historical motion information of all targets, lane topology in high-precision maps, and traffic rules into the decision to derive a judgment on behavioral intent. In recent years, the rise of end-to-end fused neural networks has enabled multimodal data to complete joint learning within the same network structure, further improving the overall real-time performance and robustness.
3) Core Algorithm
After sensor fusion, tracking and predicting various targets becomes the next challenge. For the correlation problem in multi-target environments, algorithms such as Joint Probabilistic Data Association (JPDA) and Multiple Hypothesis Tracking (MHT) can effectively solve mismatches; while Extended Kalman Filter (EKF) and Unscented Kalman Filter (UKF) utilize vehicle dynamics models to accurately estimate the target's state. When determining "where the target will go," prediction models based on Long Short-Term Memory (LSTM) networks or Graph Neural Networks (GNNs), combined with lane information, traffic signals, and the target's turn signal status, can provide reasonable trajectory inferences within just one or two seconds. For pedestrians and cyclists, the system further analyzes key human features, identifying head orientation and walking posture to determine if the pedestrian intends to cross the road. Throughout this process, data latency control and packet loss recovery are also crucial; otherwise, outdated decisions may occur, or even necessary warnings may not be triggered.
4) System architecture and software platform
The high performance and reliability of active safety driver assistance systems rely on a well-defined, real-time responsive, and redundant electronic/electrical (E/E) architecture. Traditional automotive electronics are mostly "distributed architectures," where each functional module (such as AEB, LKA, BSD) corresponds to an independent ECU (electronic control unit), communicating via protocols such as CAN bus or FlexRay. While this approach offers a clear structure and independent modules, it suffers from limitations in communication bandwidth, redundant control logic, and low coordination efficiency, especially when multiple functions are integrated, resulting in insufficient real-time response.
Therefore, more and more OEMs are transitioning to a "centralized architecture," which integrates multiple active safety and driver assistance functions into one or more high-performance domain controllers (ADCs or Zone Controllers), unifying the algorithm execution and centrally scheduling perception and control signals. Core functions such as automatic emergency braking, lane keeping, and adaptive cruise control can be integrated into a central computing platform for active safety and driver assistance. Multiple heterogeneous cores (such as NPU+CPU+DSP) execute the perception, decision-making, and control processes respectively, thereby significantly improving response speed and reducing ECU hardware costs.
On the software platform side, active safety and driver assistance functions are typically deployed in operating systems compliant with the AUTOSAR (Automotive Open System Architecture) standard and encapsulated using a service-oriented architecture (SOA) to allow different functions to communicate through interface specifications. Mainstream perception algorithms mostly run in Linux or QNX environments, while the real-time control portion uses an RTOS to ensure millisecond-level response requirements. Autonomous driving chips (such as NVIDIA Orin, Mobileye EyeQ5, and Huawei MDC) provide rich acceleration libraries supporting real-time inference of convolutional neural networks, radar point cloud preprocessing, and trajectory planning algorithms, becoming the core computing power guarantee for modern active safety and driver assistance platforms.
5) Functional safety and redundancy design
In active safety systems, every decision can directly impact driving safety; therefore, functional safety design is considered the lifeline for technology implementation. Currently, the industry generally follows the ISO 26262 standard for systematic functional safety assessment, requiring layer-by-layer verification at the system, hardware, and software levels to ensure that each function will not lead to loss of control in the event of a failure. Critical modules (such as AEB or LKA) need to have their ASIL (Automotive Safety Integrity Level) assessed, ranging from A to D, with ASILD being the highest level, requiring redundant computing paths, redundant power supplies, and redundant actuators.
6) Data-driven and self-learning systems
Many functions of traditional active safety driver assistance systems are primarily driven by explicit rules, such as determining lane departure based on lane line geometry models and assessing collision risk based on TTC time windows. This approach performs well in highway environments with clear rules and stable data, but it is prone to failure in urban roads, congested traffic, or unstructured scenarios (such as construction, rain, or snow cover).
Therefore, in recent years, an increasing number of active safety driver assistance systems have adopted a "data-driven" modeling approach. For example, in pedestrian behavior prediction, the system no longer judges danger solely based on distance and direction, but instead uses deep learning models to model information such as the pedestrian's historical trajectory, body posture, and line of sight, thereby predicting their potential behavior in the next 2-3 seconds. In lane change assist and following acceleration control, the system is also gradually moving away from traditional PID rules and towards data-trained reinforcement learning or imitation learning controllers to achieve more natural outputs that more closely resemble human driving styles.
"Automatic data refeeding" and "closed-loop learning" have become catalysts for the rapid evolution of active safety-assisted driving technologies. During each test or real-world driving session, the system automatically labels key scenarios (near-miss events, false alarms/misjudgments, extreme weather) and uploads them to the cloud for subsequent model optimization. By building high-quality data platforms, automatic labeling systems, and model training pipelines, OEMs and suppliers have constructed a closed-loop link from mass-produced vehicles to the training platform, enabling the continuous evolution of active safety-assisted driving systems.
Review of Active Safety and Driver Assistance Features
1) Automatic emergency braking
Automatic Emergency Braking (AEB) is one of the earliest mass-produced features in active safety-assisted driving and represents the most representative scenario of this technology. In each calculation cycle, the system calculates Time-to-Collision (TTC) and Braking-to-Collision (BTC) in parallel, combining the vehicle's braking performance curve and road friction coefficient model to determine whether a safe stop can be achieved within the remaining distance. When a risk is detected to be beyond the controllable range, and the driver fails to apply the brake pedal in time, the Vehicle Electronic Control Unit (VECU) prioritizes issuing braking intervention commands and invokes the ABS and Electronic Stability Control (ESP) subsystems to achieve optimal braking force distribution. The entire process must be completed within tens of milliseconds, requiring the braking assist system and brake sensors to exhibit extremely high consistency and reliability.
2) Forward Collision Warning
Before AEB (Autonomous Emergency Braking) is activated, Forward Collision Warning (FCW) serves to alert the driver. The system calculates a collision risk index in real time using fused target tracking results and alerts the driver to apply the brakes or steer when the TTC (Traffic Time Tolerance) reaches a certain warning threshold (e.g., 1.5 seconds). FCW emphasizes "giving the driver time to react." By monitoring the distance and speed of the target in real time, it can issue a warning in the early stages of an accident, fundamentally reducing the frequency of AEB activation and mitigating the braking impact on the driver.
3) Adaptive cruise control
Adaptive Cruise Control (ACC) is one of the core control modules for achieving semi-autonomous driving. Its goal is to allow the vehicle to automatically follow the vehicle in front without driver intervention, maintaining a set speed or safe distance, and accelerating or decelerating dynamically according to traffic flow changes. ACC needs to accurately detect the relative distance and speed between the vehicle and the vehicle in front. This is typically achieved using millimeter-wave radar, which calculates Doppler information and distance contours of obstacles ahead using the FMCW signal structure, and then combines this with target classification logic to eliminate interference from non-vehicle objects such as road signs and bridges.
Once the perception layer confirms the presence of the target vehicle, the ACC decision module calculates a safe timeheadway, typically based on 1.5–2 seconds, and predicts the optimal target speed for the vehicle based on the target vehicle's acceleration trend and road conditions. At the control level, the vehicle executes a longitudinal controller based on Model Predictive Control (MPC) or adaptive PID algorithms. This controller considers factors such as current speed, target speed, vehicle mass, gradient, and braking delay, smoothly adjusting throttle opening and braking force output to minimize passenger discomfort. On the intelligent driving chip, this type of controller uses high-speed inference capabilities of real-time data streams to provide feedback within 20ms–50ms, ensuring stable following behavior in both high-speed and congested conditions.
While Adaptive Cruise Control (ACC) seems capable of handling most driving scenarios, it can misjudge situations at low speeds in urban areas, such as a vehicle suddenly stopping, a cyclist cutting in, or a traffic signal recognition failure. To address this, some manufacturers are deeply integrating ACC with camera-based perception systems to improve its robustness in complex environments.
4) Lane keeping and lane departure warning
Another widely deployed active safety feature is Lane Keeping Assist (LKA) and Lane Departure Warning (LDW). These two features primarily use cameras to detect the shape, type, and position of lane lines in real time, determine whether the vehicle has deviated from its lane, and provide steering intervention or warnings accordingly.
In terms of algorithm implementation, the camera first undergoes image distortion correction and enhancement processing, and then extracts the boundary features of the road lane lines using a deep neural network (such as SCNN or ENet). These features are then mapped to the vehicle coordinate system to construct a lane model. Currently, quadratic curves are often used to fit the lane lines, and the lateral deviation and angular error between the vehicle and the lane centerline are estimated using camera pose and vehicle IMU data. When the lateral deviation exceeds a certain threshold, the system will activate a warning, providing audible or vibration feedback to the driver. If equipped with LKA (Lane Assist Key) functionality, a slight corrective torque will be applied using the electric power steering (EPS) system to help the vehicle return to the vicinity of the lane centerline.
More advanced versions, such as Lane Centering Control (LCC) and Highway Assist, add tracking of the preceding vehicle's trajectory, combining lane geometry and vehicle dynamics models to achieve precise trajectory following even on curved roads. Modern vehicles' LKA modules also incorporate robust fault-tolerance mechanisms; for example, when road markings are severely worn or obscured, the system reduces intervention intensity to avoid erroneous corrections that could have adverse effects.
5) Blind spot monitoring and lane change assist
Blind Spot Detection (BSD) and Lane Change Assist (LCA) primarily address safety issues in blind spots to the sides and rear. This functionality is mainly achieved through a 24GHz millimeter-wave radar mounted at the rear corner. This type of radar has a wide horizontal field of view and a medium detection range, making it suitable for monitoring target vehicles traveling in adjacent lanes or rapidly approaching them.
Technically, the BSD system continuously tracks the space approximately 3–5 meters to the side and rear of the vehicle, analyzing the target vehicle's movement trend. If the target remains in this area for an extended period, the system illuminates a warning symbol via the exterior rearview mirror or instrument panel icon; if the target is still in the blind spot when the driver activates the turn signal, the system triggers a stronger audible or haptic warning. Some advanced versions of LCA also actively suppress lane change maneuvers, using steering resistance feedback or a short delay in driver input to avoid collisions.
Working in conjunction with BSD is Rear Cross Traffic Alert (RCTA), which mainly operates when a vehicle is reversing. It uses lateral radar to detect moving targets in the horizontal path, such as pedestrians or vehicles crossing the road, and issues braking or audible and visual warnings to prevent reversing accidents.
6) Traffic sign recognition and speed limit assist
With the maturity of computer vision technology, vehicles have begun to have the ability to recognize traffic signs, especially in speed limit recognition and violation sign recognition, where relatively mature mass-production solutions have been developed. The system mainly relies on a forward-facing camera and uses a combination of OCR (Optical Character Recognition) and convolutional networks to detect and parse the pattern, number, and color information in traffic signs in images.
In the recognition process, the system first extracts the edge and shape features of the image, filtering typical circular, triangular, and octagonal regions. Then, it performs character segmentation and classification to identify specific sign content such as "Speed Limit 60," "No Left Turn," and "School Zone." Some technical solutions also integrate the results from high-precision maps and V2X communication modules to double-verify the recognition results, thereby improving accuracy. When a new speed limit area is detected, the system can proactively adjust the target speed of adaptive cruise control or issue a warning when the driver exceeds the speed limit.
7) Driver monitoring
Ensuring the driver remains alert at all times is a prerequisite for the safe operation of advanced driver assistance systems (ADAS). Driver Monitoring System (DMS) technology has therefore become a key focus. This system typically consists of an infrared camera or a Time-of-Flight (TOF) camera mounted near the steering wheel or dashboard to continuously analyze the driver's eye movements, head posture, and facial expressions.
Using facial landmark extraction algorithms and eye-tracking models, the system can identify whether the driver is looking ahead, whether they have closed their eyes for a certain period of time (drowsy), and whether they are frequently looking down (looking at a mobile phone), which is considered dangerous behavior. In some models, the system also monitors facial temperature and skin texture changes to determine the driver's level of fatigue or abnormal alcohol intake. When a potential risk of loss of control is detected, the system can activate warning lights, steering wheel vibration, voice prompts, and even trigger AEB or low-speed braking functions at different stages to ensure the continuity and safety of vehicle operation.
Final words
Although current active safety assistance functions are still classified as Level 1/L2, they form the cornerstone for the transition to Level 3/L4 autonomous driving. In Level 3 scenarios (such as automatic lane changing, high-speed ramp entry and exit, and intelligent lane switching on highways), the requirements for environmental modeling, behavior prediction, and system stability are significantly increased. Level 2 active safety assistance focuses on "assisting humans," while systems at Level 3 and above must have the ability to "take over decision-making," meaning they cannot rely solely on rules or triggers but must possess a complete scenario understanding capability and a highly reliable behavior generation system.
In this transformation, the perception range and control capabilities of active safety systems are expanding. From AEB to pedestrian avoidance at intersections, from LKA to automatic lane centering and following in urban lanes, from FCW to traffic light recognition and priority judgment in complex traffic situations. This integrated development means that future intelligent driving systems will no longer distinguish between "active safety" and "autonomous driving," but will instead merge into a unified intelligent driving stack, layered by capabilities rather than by functions.
In the future, with the continuous maturation of algorithms such as image recognition, point cloud modeling, and graph neural networks, coupled with the leapfrog development of hardware computing power, active safety systems will shift from passive response to proactive understanding, from rule enforcement to strategy generation, and ultimately become the core "brain" of the vehicle's perception-understanding-action trinity. At the same time, it will also provide a stable and reliable safety barrier for L3+ level autonomous driving functions, both safeguarding the bottom line and illuminating the bright path to fully autonomous driving.
