The proposed paper presents a feasibility study on Time-of-Flight (ToF) cameras for Reverse Autonomous Emergency Braking systems (reverse AEB) and the application of computer vision techniques for automatic target identification in the real-time.
The consumer testing organisations play an important role in the expansion of vehicle safety systems. The New Car Assessment Programme (NCAP) organisations assess vehicle safety considering different areas, for instance in Europe the rating of EuroNCAP consists of four categories: Adult Occupants, Child Occupants, Vulnerable Road Users (VRU) and Safety Assists. One of the tested systems is Autonomous Emergency Braking (AEB), which activates automatically the vehicle's brakes in order to prevent or reduce the severity of a collision. The new EuroNCAP protocol, effective from January 2020, brought the application of AEB also to reverse motion of vehicles to prevent accidents with pedestrians. The Reverse AEB automatically activates brakes when a potential obstacle behind the vehicle is detected. The system has the ability to reduce a significant number of accidents, particularly during parking manoeuvres.
The proposed work aims to investigate the suitability of the innovative PMD technology with a ToF Camera for the reverse AEB. Following the protocol guidelines for the assessment test conditions, the Vehicle Under Test (VUT) should be tested with speeds of -4 and -8 km/h with impact offsets of 25, 50 and 75%. The protocol considers an adult pedestrian target.
However, the presented research goes beyond the requirements of the protocol and addresses the case of a certified child target, which is more dangerous because of its smaller size. The work consists of a data acquisition phase with a real ToF camera sensor in a scenario according to the guidelines of the AEB VRU Systems protocol from EuroNCAP. The camera generates point clouds information in 3D format, which is acquired at a rate of 10 FPS (frames per second). The captured 3D data are stored and used as an input for the development in MATLAB/Simulink, where the identification algorithm is implemented. It locates the child target's position and determines the distance relative to the vehicle. Further, a decision-making system decides to activate the brakes according to their limits, avoiding collision with the target on one hand side and reducing false positive issues on the other hand side. In order to simulate the braking manoeuvre, the vehicle dynamic model is developed in a MATLAB/Simulink environment. Finally, the paper summarizes the advantages and disadvantages of the application of the proposed technology.
Mr. Rodrigo Negri de Azeredo, Federal University of Parana - UFPR, BRAZIL; Prof. Dr. Ondrej Vaculin, Technische Hochschule Ingolstadt - THI, GERMANY; Prof. Dr. Gustavo Henrique da Costa Oliveira, Federal University of Parana - UFPR, BRAZIL