Analysis of AWS Rekognition and Azure Custom Vision Performance in Parking Sign Recognition

Automated recognition and analysis of parking signs can greatly enhance the safety and efficiency of both autonomous vehicles and drivers seeking navigational assistance. Our study focused on identifying parking constraints from the parking signs. It offers the following novel contributions: (1) A comparative performance analysis of AWS Rekognition and Azure Custom Vision (CV), two leading services for image recognition and analysis. (2) The first AI-based approach to recognising parking signs typical for Melbourne, Australia, and extracting parking constraint information from them. We utilised 1225 images of the parking signs to evaluate the AI capabilities for analysing these constraints. Both platforms were assessed based on several criteria, including their accuracy in recognising elements of parking signs, sub-signs, and the completeness of the signs. Our experiments demonstrated that both platforms performed effectively and are close to being ready for live application on parking sign analysis. AWS Rekognition demonstrated better results for recognition of parking sign elements and sub-signs (F1 scores of 0.991 and 1.000). It also performed better in the criterion "No text missed", providing the result of 0.94. Azure CV performed better in the recognition of arrows (F1 score of 0.941). Both approaches demonstrated a similar level of performance for other criteria.