Deep Learning-Based Parking Detection System Using Structural Similarity
High latency of existing vision-based parking detection systems causes significant delay in parking information delivery. The lack of access to real-time outdoor parking information induces the difficulty of finding parking and identifying parking violations, especially in busy downtown areas. Given...
Saved in:
| Main Author: | |
|---|---|
| Format: | Thesis |
| Published: |
2019
|
| Subjects: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | High latency of existing vision-based parking detection systems causes significant delay in parking information delivery. The lack of access to real-time outdoor parking information induces the difficulty of finding parking and identifying parking violations, especially in busy downtown areas. Given such backdrop, this research project put forward a deep learning-based parking detection system that accurately detects parking occupancy status and illegal parking on real time basis. A Convolutional Neural Network (CNN) classifier running on a single board computer is used to classify the occupancy status of parking spaces and identify the presence of illegally parked vehicles. A novel structural similarity (SSIM) decision scheme is introduced to minimize the latency of detecting parking occupancy changes. Multithreading technique is also employed to fully exploit the processing power of single board computer for accelerating illegal parking detection process. The detection results are stored at a cloud-hosted NoSQL database. Two smartphone applications namely Driver App and Enforcer App are developed to channel live parking information for assisting drivers in finding parking and supporting efficient parking enforcement respectively. The proposed parking detection mechanism is implemented and tested at the staff parking area (monitored with an IP camera) of Faculty of Engineering in Multimedia University. Experimental results demonstrated a high accuracy of over 99% in parking occupancy detection and 98.7% precision rate in locating illegal parking incidents with no missed detection under different weather conditions. The implementation of SSIM decision scheme in detecting a single instance of parking occupancy change, managed to accelerate the parking occupancy detection by up to six times faster when compared to a pure CNN classification approach. On the other hand, the incorporation of multithreading mechanism on single board computer increases the computation speed of illegal parking detection in three illegal parking regions by 2.3 times in comparison with sequential execution. |
|---|