| Title |
Method of Real-time Indoor Localization and Monitoring in Parking Lot using Image Deep Learning Analysis |
| Authors |
백장현(Jang-hyun Baek) ; 노용철(Yong-cheol Ro) ; 장준혁(Jun-hyek Jang) ; 장수현(Soo-hyun Jang) ; 김진규(Jin-kyu Kim) |
| DOI |
https://doi.org/10.5573/ieie.2023.60.11.86 |
| Keywords |
Indoor localization; Parkinglot monitoring system; Re-identification; Deep learning |
| Abstract |
This paper introduce an effective method for indoor localization in an indoor parking lot and monitoring in real time by analyzing surveillance camera images installed in a parking lot. Proposed method analyzes the camera image installed in the parking lot using a deep learning to express the location of the vehicle as integrated grid map coordinate, which can express the location of all vehicles in the parking lot in real time. The method consists of three deep learning models: object detection, tracking and license plate recognition and coordinate transformation algorithms. The deep learning model was optimized by applying quantization of NVIDIA TensorRT, and all camera video input was organized into one process using Deepstream to increase the efficiency of hardware resource usage. Unlike conventional indoor positioning technologies that require sensors to be installed in parking lots or vehicles, the proposed method is economical because it does not need additional sensors. 14,000 image collected from parking lots were used for deep learning model learning. The experiment conducted with 120 images showed 98.1% Re-ID performance and the location estimation error was 3.65m on average at a distance of 40m. |