No abstract

No abstract

No abstract

We present a method for detecting parking spaces in radar images based on convolutional neural networks (CNN). A multiple-input multiple-output radar is used to render a slant-range image of the parking scenario and a background estimation technique is applied to reduce the impact of dynamic interference from the surroundings by separating the static background from moving objects in the scene. A CNN architecture, that also incorporates mechanisms to generalize the model to new scenarios, is proposed to determine the occupancy of the parking spaces in the static radar images. The experimental results show very high accuracy even in scenarios where little or no training data is available, proving the viability of the proposed approach for its implementation at large scale with reduced deployment efforts.

Classification of road users is important for traffic monitoring. The usability of a height estimate based on the tworay ground-reflection model as a feature for the classification of vehicles is analyzed in this paper. The four-ray ground-reflection model for fast chirp ramp sequence waveforms of FMCW radars is derived and simplified to the well-known two-ray groundreflection model. A spectrum from which the height of a target can be derived is obtained using the Lomb-Scargle periodogram. Measurements with two vehicle classes illustrate the approach and show that the model could be used as a feature to distinguish vehicles based on their height.