Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Vehicle Occurrence-based Parking Space Detection

Paulo R. Lisboa de Almeida, Jeovane Honório Alves, Luiz S. Oliveira, Andre Gustavo Hochuli, João V. Fröhlich, Rodrigo A. Krauel

TL;DR

This work tackles automatic parking space detection from image sequences without relying on manually labeled spots. It presents a pipeline that uses Cascade Mask R-CNN with a ConvNeXt-S backbone to segment cars, builds a heat map of car occurrence over a day, and extracts rotated-rectangle parking-space coordinates via an a posteriori scoring approach. On twelve PKLot and CNRPark-EXT subsets, the method achieves up to $AP_{25}=95.60\%$ and $AP_{50}=79.90\%$ for parking-space detection, with broader trends indicating strong performance under dataset shifts but sensitivity to illegal parking. The approach offers a practical, labeling-free solution suitable for smart parking deployments and enforcement, capable of adapting to dynamic parking environments without relying on pre-labeled spot demarcations.

Abstract

Smart-parking solutions use sensors, cameras, and data analysis to improve parking efficiency and reduce traffic congestion. Computer vision-based methods have been used extensively in recent years to tackle the problem of parking lot management, but most of the works assume that the parking spots are manually labeled, impacting the cost and feasibility of deployment. To fill this gap, this work presents an automatic parking space detection method, which receives a sequence of images of a parking lot and returns a list of coordinates identifying the detected parking spaces. The proposed method employs instance segmentation to identify cars and, using vehicle occurrence, generate a heat map of parking spaces. The results using twelve different subsets from the PKLot and CNRPark-EXT parking lot datasets show that the method achieved an AP25 score up to 95.60\% and AP50 score up to 79.90\%.

Vehicle Occurrence-based Parking Space Detection

TL;DR

This work tackles automatic parking space detection from image sequences without relying on manually labeled spots. It presents a pipeline that uses Cascade Mask R-CNN with a ConvNeXt-S backbone to segment cars, builds a heat map of car occurrence over a day, and extracts rotated-rectangle parking-space coordinates via an a posteriori scoring approach. On twelve PKLot and CNRPark-EXT subsets, the method achieves up to and for parking-space detection, with broader trends indicating strong performance under dataset shifts but sensitivity to illegal parking. The approach offers a practical, labeling-free solution suitable for smart parking deployments and enforcement, capable of adapting to dynamic parking environments without relying on pre-labeled spot demarcations.

Abstract

Smart-parking solutions use sensors, cameras, and data analysis to improve parking efficiency and reduce traffic congestion. Computer vision-based methods have been used extensively in recent years to tackle the problem of parking lot management, but most of the works assume that the parking spots are manually labeled, impacting the cost and feasibility of deployment. To fill this gap, this work presents an automatic parking space detection method, which receives a sequence of images of a parking lot and returns a list of coordinates identifying the detected parking spaces. The proposed method employs instance segmentation to identify cars and, using vehicle occurrence, generate a heat map of parking spaces. The results using twelve different subsets from the PKLot and CNRPark-EXT parking lot datasets show that the method achieved an AP25 score up to 95.60\% and AP50 score up to 79.90\%.
Paper Structure (9 sections, 6 figures, 2 tables, 1 algorithm)

This paper contains 9 sections, 6 figures, 2 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Works
  3. Proposed Method
  4. Experiments and results
  5. Datasets
  6. Experiment Settings
  7. Car Instance Segmentation
  8. Parking Space Detection
  9. Conclusion

Figures (6)

  • Figure 1: Parking space locations (image from CNRPark-EXT).
  • Figure 2: Proposed method
  • Figure 3: Examples of cameras used in our experiments: (a) UFPR05 from PKLot and (b) Camera3 from CNRPark-EXT
  • Figure 4: Automatic Parking Space Detection from Camera3 (upper) and UFPR05 (lower part): Resulting heat map in (a,d). Automatic demarcation in (b,e), where TPs are in blue, FPs in orange, and FNs in black. Human annotation in (c,f).
  • Figure 5: Worst (\ref{['subfig:worstResUFPR05']}) and (\ref{['subfig:bestResUFPR05']}) results achieved in the UFPR05 subset in the DATASET CHANGE evaluation. In (\ref{['subfig:resultByDayUFPR05']}) we show the AP25 (mean and standard deviation) achieved for every day tested. The best and worst days are highlighted in green.
  • ...and 1 more figures