AI-driven pedestrian detection systems are being developed as a response to the critical global issue of pedestrian fatalities, which have risen in some regions despite overall vehicle safety improvements. These systems integrate Artificial Intelligence into vehicles to help them perceive, understand, and react to pedestrians, often more quickly than a human driver. The core technology relies on Computer Vision to process visual data, Machine Learning and Deep Learning models (like CNNs and YOLO) trained on extensive datasets to recognize pedestrian patterns. To build a comprehensive view, Sensor Fusion combines data from multiple sensors: Cameras for visual detail, LiDAR for precise 3D mapping, Radar for range and velocity in challenging weather, and Thermal Imaging for detecting heat signatures, especially in low light. This fusion enhances object detection, classification, and helps in predicting pedestrian behavior to enable real-time decision-making, including alerts and Automatic Emergency Braking (P-AEB). Leading manufacturers and technology providers like Tesla, Volvo, Subaru, Toyota, Ford, Mobileye, Nvidia, Bosch, and Continental are implementing varying levels of these capabilities. While statistical data indicates that P-AEB systems contribute to reducing pedestrian crashes, their real-world effectiveness is highly variable. A significant limitation is performance in low-light or nighttime conditions, where some systems have been found to be "nearly ineffective", despite most pedestrian fatalities occurring at night. Adverse weather and complex urban environments also pose challenges. Current systems can struggle with occlusion, detecting small or unpredictable targets, and are susceptible to sensor obstruction. Key hurdles include achieving consistent accuracy and reliability across all scenarios and managing the precision-recall trade-off between avoiding false positives (unnecessary braking) and preventing catastrophic false negatives (missed detections). A major concern is algorithmic bias, with research showing lower detection accuracy for children and pedestrians with darker skin tones, often linked to biased training data. The inherent unpredictability of human behavior (e.g., jaywalking, sudden movements) also remains difficult for AI to reliably predict. The deployment of AI also raises significant ethical and societal questions, such as determining accountability and liability in case of system failure, addressing data privacy concerns from mass surveillance, and grappling with pre-programmed ethical choices in unavoidable accidents ("trolley problem"). Regulatory bodies like NHTSA (mandating P-AEB including nighttime performance by 2029) and Euro NCAP (with influential testing protocols) are pushing for improved pedestrian safety standards. Industry standards like ISO 26262 (Functional Safety), ISO 21448 (Safety of the Intended Functionality - SOTIF for performance limitations), and the new ISO/PAS 8800 (Safety and AI) provide guidance for developing these complex systems safely. Rigorous testing involves simulation, physical tests with soft targets, and real-world data analysis. Future trends include advanced multi-modal fusion, more accurate predictive AI, V2X communication, emerging sensors like thermal and neuromorphic cameras, and the use of Explainable AI (XAI) for transparency. In summary, while AI has undeniably made cars "smarter" and contributed to reducing pedestrian crashes, achieving genuinely "safer streets" requires overcoming significant technological limitations, particularly in challenging environmental conditions and for vulnerable demographics, alongside vigilant ethical consideration, adaptive regulation, and building public trust.
