A Parking Space Detection Algorithm Based on Semantic Segmentation

In order to solve the problem that the current parking space detection relies on the fusion of ultrasonic sensors and millimeter wave radar multi-sensors, a parking space detection algorithm based on semantic segmentation is proposed. This algorithm only relies on the image of the vehicle's around view monitor system to complete the parking space detection. The algorithm mainly includes two modules: semantic segmentation and parking space inference. Firstly, we designed a semantic segmentation model with excellent performance based on DeepLab V3 +. The image of the vehicle's around view monitor system was used as the input, and the output was the semantic segmentation results of the free spaces, slot markings, vehicles and other obstacles. After the trained semantic segmentation model, the test results on the test dataset are satisfactory, with an overall accuracy of 97.37% and a recall rate of 97.21%. The segmentation accuracy in the free spaces reached 98.22%; the segmentation accuracy of the slot markings reached 83.97%; the segmentation accuracy of the vehicles reached 93.80%; and the segmentation accuracy of other obstacles reached 89.44%. Secondly, we propose a parking space inference algorithm based on vertical grid search, using the results of semantic segmentation to infer available parking spaces. It can be seen from the test results that the available parking spaces can also be accurately inferred under difficult-to-identify scenes.