Interval-based collision prediction utilizing nested Kalman filters in video surveillance systems

Traffic safety, pedestrian safety, and collision prediction systems are of high importance, especially with the increasing demand for autonomous driving. Expansion of cities and population growth necessitate the utilization of automated video surveillance systems and computer vision-based analyzes to enhance overall traffic safety. Over the past years, several works have employed algorithms to work closely with the overall electronic architecture of the car. There are several obstacles, such as non-unified electronic architectures, accessibility of standardized hardware, regional differentiation, and legal aspects, slowing down the continuous evolution of self-driving systems with enhanced safety systems. In this work, we introduce a method for improving the prediction of collision between two objects by utilizing stationary video capture systems that comply with standardized decision-making processes, support easy-to-deploy communication protocols, allow hardware flexibility, and enable communication between vehicles, traffic lights, and roadside detection systems through software- and hardware-independent evaluation models.