Distance Estimation using Deep Learning Approaches for Rear-end Collision Avoidance Alerts

Autonomous Emergency Braking (AEB) and Autonomous Emergency Steering (AES) are part of the advanced driver assistance system (ADAS) equipped in intelligent vehicles. AEB is a system that warns drivers of potential collisions and assists them in utilizing the vehicle's maximum capabilities. AES is an active safety system that aids in evasive steering. If it detects a potential collision, unlike AEB, the AES system will autonomously adjust the steering to prevent it. The challenges for AEB and AES include determining how much space is required to avoid an accident while turning or braking and how much distance is required to avoid an impact when braking and turning simultaneously. Considering such inquiries, it is necessary to devise a system to estimate the distance between the vehicles. Therefore, this study proposes a Monocular Vision Distance Estimation (MVDE) method employing deep learning techniques for accurately calculating the distance between vehicles, particularly for use in AEB and AES systems. The MVDE technique uses monocular vision, emphasizing object detection and distance estimation. In contrast to complex depth estimation techniques, the proposed method employs a Single Shot Detector (SSD) with MobileNet architecture for object recognition and Deep Artificial Neural Networks (Deep ANN) for accurate distance estimation. Using a real-world dataset collected in Cyberjaya, Malaysia, this study rigorously assesses the performance of this method. Results indicate that the MVDE method with four hidden layers in Deep ANN outperforms earlier techniques, with a maximum measured error of 4m to actual distances. In addition, it is competitive with RADAR-based systems and offers a cost-effective alternative for widespread adoption. These findings support the potential of MVDE for augmenting vehicle safety, shaping future automotive standards, and facilitating the widespread implementation of AEB and AES systems.