When I didn’t find a parking space at work, the hard-core programmer made a "real-time parking space recommendation system in the park" and won an award.

At the annual 1024 Programmer's Day, the NetEase Technology Marathon kicked off again. In a limited time of 48 hours, 16 teams and a total of 84 technology lovers competed.

Unlike most projects and businesses that are closely related, there is a team whose projects are closer to real life. for Parking Spaces” team, the participating topic is 161974b8258956 "Park Parking Space Real-time Recommendation System" , applying visual recognition, speech synthesis and other technologies to create a parking solution. Finally, in this technical marathon, he won the Technical Geek Excellence Award.

Speaking of why such a topic was chosen, team leader Shinji said, “Sometimes it takes a lot of time to park for work, and parking is a very common problem at present, so I want to design a solution with technology.”

1. Project analysis

Everyone with a car should have had the fear of being dominated by parking. Whether it is in a shopping mall or a park, it clearly shows that there are still vacancies, but after two rounds, I don’t know where... I finally found a vacancy, but I was about to drive. In the past, it was robbed first...

To sum up, there are two difficulties-finding a parking space and grabbing a parking space.

So, how to use technology to solve these two problems?

The overall design idea is:

Relying on the video vehicle detection algorithm, it achieves real-time monitoring of multiple parking spaces, flexibly adapts to the effective area and reduces the algorithm's misrecognition rate to a certain extent. Use effective information such as vehicles and owners in the work park to reasonably allocate parking areas for incoming vehicles, reduce the cost of manual parking commands, and improve parking efficiency.

Two, the solution

1. Parking area monitoring

Use the vehicle recognition algorithm to monitor the vehicles within the monitoring range, and report the data in real time.

This method of construction is less difficult, can monitor multiple parking spaces, and can filter out invalid areas. However, it is more dependent on the accuracy of the recognition algorithm, and there are certain requirements for the installation position and angle of the camera.

For vehicle detection, we have considered two options for actual scenarios.

• Solution 1: Based on deep learning and image recognition technology, edge computing is used, and the camera has a built-in algorithm for vehicle detection.
• Solution 2: The camera only collects images and uploads them to the server every 10 seconds. The server then calculates vehicle detection on the data.

The advantage of scheme one compared with scheme two is that the camera recognizes the number and specific positions of vehicles in the current shooting range, and notifies the server when the number changes. Edge computing will greatly reduce the bandwidth and computing pressure of the server. But the problem is that the cost of this camera will be relatively high.

However, both the first and second options have certain requirements for the placement of the cameras. A camera should clearly cover as many parking spaces as possible. The ideal situation is to cover 10-12 parking spaces, and the specifics should be combined with the actual situation. The main factor in covering enough parking spaces is to reduce the cost of hardware, that is, the cost of the camera.

Based on the company's existing camera, the initial model we used is Option 2.

2. Division of parking spaces

Divide all parking spaces in the parking lot into several areas according to conditions such as location, size, and monitoring area.

The minimum granularity of management is expanded from a certain parking space to a group of parking spaces, which facilitates route planning and reduces management difficulty.
When assigning seats, car owners are given certain options to adapt to the complex needs in the assignment and are conducive to promotion.

3. Recommended parking spaces

• The vehicle enters the venue, and the barriers recognize the license plate or vehicle model data and report the recommended service.

• Recommend service to match vehicle information.

  • If the matching is successful, all the parking areas are roughly ranked according to the number of free parking spaces, and then rearranged according to the matching strategy, and finally the recommended parking spaces are confirmed according to the diversion strategy;
  • If the match fails, confirm the recommended parking space according to the default maximum vacancy rule.
• Pre-occupied parking spaces will enter the regional pre-occupation pool. After the expiration date, the pre-emption will be invalidated.
• The recommended parking space will be informed to the gate and APP, and a voice announcement or notification will be sent to inform the owner of the parking area.

The recommended parking space will consider more than

• The parking space is away from the stairs, the location of the elevator, and further the stairs, elevators and the specific work positions of the employees.
• According to the route of the vehicles that have entered, the locations of vehicles that are not particularly large are recommended.
• Stay away from luxury cars.
• Employees are used to parking locations.
• New energy first recommends rechargeable parking spaces for new energy vehicles.
• Large parking spaces are recommended for large cars first to reduce the chance of small cars using large parking spaces.
• Further open up the conference system. If the meeting time is relatively close, recommend the most reasonable parking space according to the location of the meeting room and the distance from the existing vacant parking spaces, which can be quickly parked. Be as individualized and rational as possible.

3. Project application effect display

https://www.bilibili.com/video/BV18Q4y1v7D7/?aid=719339220&cid=444998568&page=1

Summarize

In order to solve the problem of finding and grabbing parking spaces in the parking lot scene, we introduce visual recognition technology and recommendation algorithms into the parking management project. Relying on the video vehicle detection algorithm, it achieves real-time monitoring of multiple parking spaces, flexibly adapts to the effective area and reduces the algorithm's misrecognition rate to a certain extent. Use effective information such as vehicles and vehicle owners in the work park to reasonably allocate parking areas for incoming vehicles, reduce the cost of manual parking commands, and improve parking efficiency.

To solve the problems in the parking process by way of allocation, combined with route planning can also avoid congestion, and if there is an opportunity later, it will also consider joining the project planning. At the same time, in the future, we hope to expand support for the management of parking spaces on the ground, and further reduce the cost of project implementation, amortize the cost of algorithm detection, and allow single-channel cameras to effectively monitor more parking spaces and increase project revenue.