Finding Potential On-street Parking Spots: An Object Detection and Segmentation Approach

The scarcity of on-street parking spaces in metropolitan areas has led to traffic congestion, frustration among drivers and resource wastage. Conventional parking management systems fall short in addressing these challenges, particularly as manual searches for parking spots contribute to an increase in traffic congestion. This research contributes to intelligent parking management by proposing a novel approach to identify potential parking spots along roads using image processing and deep learning techniques. Our method includes three models and proposes a solution for identifying parking spaces in urban environments. You Only Look Once (Yolov8), a Convolutional Neural Network (CNN)-based image processing model serves as the base for three models. The three models are sequential and are used to perform different set of tasks like detect parking signs, lane lines, road edges, footpath and differentiate between parked and moving vehicles. The models’ outputs also act as an elimination criterion at each step thereby focusing on predicting a potential parking spot. The experimental analysis is performed in terms of F1-scores that provide 0.96 at 0.96 IoU, 0.39 at 0.186 IoU, 0.53 at 0.315 IoU for the first, second and third models, respectively.