Vision-based Parking Lots Management Technology for Smart Parking

Bei Jing, August 18, 2015

Fujitsu Research & Development Center Co., Ltd. (Note 1) has developed a vision-based parking lots management technology for smart parking. Proposed technology relies on surveillance videos to analyze status of each parking lot and provides detailed occupancy maps to improve convenience for both administrator and drivers.

Compared with existing methods and products, our technology is applicable for various scenes, including outdoor and indoor parking sites, and able to cover up to 100 parking lots by one camera efficiently. Besides, with patent-pending image analyzing technique, our technology can achieve high detection precision even under all-weather conditions.

【Background】

As the number of private vehicles increases rapidly in these years, parking has become a serious problem for both drivers and the government. For drivers, it usually takes long time to find a valid parking lot; for authorities, the statistical information of public and chargeable parking lots will help make better planning to relieve the problem. Therefore, a smart parking solution is expected to elevate the parking experience by checking availability of each parking lot at real-time, guiding drivers effortlessly to a vacant parking space and providing statistical data in advance for cost cutting.

As the critical part of smart parking solution, the technology to detect occupancy status of a parking lot has developed through three stages: Inductive Loops based (1970~1980s), Ultrasonic Sensors based (1990s), Camera based (2010s). Camera based smart parking solutions have been increasing rapidly with popularization of large scale video surveillance systems, as they provide more video information and able to support advanced functions.

【Topics】

Although smart parking solutions based on all technologies mentioned above are still deployed in difference scenes, there are problems or constraints existing. To be specific:

Inductive loops based technology and ultrasonic sensors based technology are most widely adopted due to low cost. However, detection precision is not satisfying enough for parking status monitoring, as well as being limited by environment conditions. For example, systems that utilize induction loops are interfered by severe weathers (e.g. lightening) and neighboring sensors; systems that utilize ultrasonic sensors are not suitable for outdoor implementation.

Camera based technology has some different implementation ways according to detecting approaches. Most existing solutions or products will use one customized camera to cover 2~3 parking lots, thus entire system will be very expensive due to hardware cost, let alone such solutions are not suitable for outdoor implementation; Some solutions are to evaluate the number of entering and exiting vehicles within a certain area instead, system cost will be reduced but detailed status information of each parking lot are not available.

【Technology】

Figure 1 illustrates the framework of proposed parking lots management technology, it’s composed of three main parts: data acquisition, data analyzing and information publishing.

Figure 1 Parking lots management technology framework

Firstly, we acquire video data from surveillance cameras. By considering planning of actual parking site and carefully deploying cameras with proper height and angle, optimized coverage efficiency could be achieved. For example, for an outdoor parking site without any roof, one 1080p camera with focal length 2.8mm and installment height 20m is able to cover 80~100 parking lots; for other situations that cover smaller area, camera installment height could be reduced correspondingly. Data storage devices and central management server are used depending on whether users demanded to store videos for review.

The pivotal analyzing process is carried by FRDC superior parking lot analyzing software on analyzing workstations. As shown in Figure 2, process will be executed on each parking lot patch. Image preprocess includes the operations of noise filtering and image enhancement, confronting bad light conditions such as nighttime and foggy weather. Change judgment is employed to avoid redundant detecting operations and improve stability of detection result, in other words, we analyze the latest occupancy status of a parking lot only when there are changes happened within the area. Besides, to boost efficiency of entire system, we bring in polling scheme to process multiple camera streams alternatively. In this way, one analyzing workstation is able to take processing load for 8 cameras or more.

Figure 2 Analyzing process flow

Figure 3 shows the effect of image preprocess for noise filtering and image enhancement.

Figure 3 Image preprocess result (Top: original images; Bottom: processed images)

Finally, detailed parking occupancy status map is obtained. Such information could be published to web server for administration or through mobile applications for drivers in the future. Historical data is possible to be synthesized to provide more valuable information, for example, predicting peak time and pricing flexibly.

【Result】

Proposed technology has taken Proof-of-concept test in a Mideast city and field trial in a southern Chinese city. Here some trial results are shown for the latter:

Time: 14:00 ~ 20:00

Scene: outdoor parking site

Implementation: 1 camera covers 5~7 parking lots

Precision: >95%

Figure 4 shows the trial scene and processed results, it needs to be noticed that plants limits installment height of cameras.

Figure 4 Trial result 【Future Plans】

In 2015, we will continue field trials with mentioned projects in several cities to improve technical precision and robustness, as well as improving and optimizing entire solution. Meanwhile, we will make efforts to push forward our technology as commercialization after trial stage, and promote the technology in both China and worldwide. Also, we consider integrating core analyzing process software in front-end camera to further reduce solution cost and implementation complexity.

【Note 1】

Fujitsu Research & Development Center Co., Ltd.: Chairman Shigeru Sasaki. Location: Beijing, China