Enhancing the Accuracy of Parking Assistant Sensors with Bayesian Filter

Abstract

Sonar distance sensors are commonly used for obstacle detection and distance measurement, providing input information for different applications, such as collision avoidance algorithms and vehicle parking assistants. However, they have a wide range of quality and accuracy, resulting in prices ranging from $0.55 to over $100 per unit. As it is often necessary to use a few units in parking assistants and those are deployed on a largescale vehicle production, the unit price is a critical factor. However, the simple choice of the lowest price sensors directly impacts on the measurements reliability, since they have high levels of noise in the values of their measurements. Therefore, this presents the results of the experiments using the Bayesian Recursive Estimation technique – also known as Bayesian Filtering – to increase the accuracy and reliability of low-cost sonar sensor measurements. A prototype is implemented and evaluated in simulated and real (physical) experimental environments. Using this approach, a significant accuracy improvement on distance measurements was observed compared to the raw data obtained from sensors. The results suggest this approach can be an alternative to be considered to reduce costs when equipping vehicles with parking assistants.