In the annual 1024 programmers festival, netease technology marathon again kicked off. Limited time 48 hours, 16 teams, a total of 84 technology enthusiasts began to compete.

Unlike most projects and businesses, one team’s project is closer to real life. The “Parking Spaces don’t have to wait” team from Youdao participated in the project of “Real-time Parking Space Recommendation System in the Park”, which created a parking solution by using visual recognition, speech synthesis and other technologies. Finally, I won the Technology Geek Excellence Award in this technology marathon. **

When asked why he chose such a project, Shinji said, “Sometimes it takes a lot of time to park for work, and parking is a common problem, so I wanted to design a solution in a technical way.”

I. Project analysis

If you have a car, you probably have the fear of being stopped. No matter in the shopping mall or the park, it clearly shows that there is an empty space, but you don’t know where it is after two laps. Finally found a vacancy, but was about to open the past was robbed first…

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

So how do you use technology to solve these two problems?

The overall design idea is:

Relying on the video vehicle detection algorithm, the real-time monitoring of multiple parking Spaces can be achieved, and the effective area can be flexibly adapted and the algorithm error recognition rate can be reduced to a certain extent. The effective information of vehicles and owners in the work park is used to reasonably allocate parking areas for incoming vehicles, so as to reduce the cost of manual parking command and improve parking efficiency.

Second, solutions

1. Parking area monitoring

The vehicle recognition algorithm is used to monitor the vehicles within the monitoring range and report the data in real time.

This method is easy to construct, can monitor multiple parking Spaces, and can screen out invalid areas. However, it depends on the accuracy of the recognition algorithm and has certain requirements for the installation position and Angle of the camera.

Vehicle detection we consider two schemes for the actual scene.

  • Scheme 1: Based on deep learning and image recognition technology, edge calculation and camera built-in algorithm are used for vehicle detection.

  • Scheme two: the camera only collects images and uploads them to the server every 10 seconds. The server then calculates the data for vehicle detection.

The advantage of scheme 1 compared with Scheme 2 is that the camera recognizes the number and specific location of vehicles in the current shooting range, and informs the server when the number changes. Edge computing greatly reduces bandwidth and server computing stress. But the problem is that the cameras can be expensive.

However, no matter plan 1 or Plan 2, there are certain requirements for the placement of cameras. One camera should clearly cover as many parking Spaces as possible, ideally covering 10-12 parking Spaces. The specific situation should be combined with the actual situation. The main factor in covering enough parking Spaces is to reduce the cost of hardware, namely the cost of cameras.

Based on the company’s existing cameras, we used plan 2 for the initial model.

2. Parking area division

All parking Spaces in the parking lot are divided into several areas according to the location, size, monitoring area and other conditions.

The minimum granularity of management is expanded from a certain parking space to a group of parking Spaces, which is convenient for route planning and reduces the difficulty of management. The car owners have certain choice rights when assigning positions, which is suitable for the complex needs of allocating and conducive to promotion.

3. Parking space recommendation

  • Vehicle admission, road brake identification license plate or model data report recommendation services.
  • Recommended services match vehicle information.
    • After successful matching, all available parking areas are roughly arranged according to the number of free parking Spaces, and then rearranged according to the matching strategy. Finally, recommended parking Spaces are confirmed according to the shunting strategy.
    • If the match fails, confirm the recommended parking space according to the default maximum space rule.
  • Pre-occupy the parking space and enter the area pre-occupy pool. Preoccupation will expire after expiration.
  • The recommended parking space will be notified to the gate and APP, and voice broadcast or notification will be sent to inform the car owner of the parking area.

Parking space recommendations take into account multiple dimensions.

  • Parking space from the location of stairs, elevator, further stairs, elevator and employee specific location.

  • Based on the route of incoming vehicles, the location of not too many vehicles is recommended.

  • Stay away from fancy cars.

  • Employees are used to parking locations.

  • New energy priority recommended new energy vehicle rechargeable parking space.

  • Large cars have priority to recommend large parking Spaces to reduce the probability of small cars using large parking Spaces.

  • Further open the conference system, if the meeting time is close, according to the location of the conference room and distance from the existing vacant parking space recommended the most reasonable, fast parking space. Personalize and rationalize as much as possible.

Iii. Project application effect display

Results show

conclusion

In order to solve the problem of finding and grabbing a parking space in the parking scene, we introduce visual recognition technology and recommendation algorithm into the parking management project. Relying on the video vehicle detection algorithm, the real-time monitoring of multiple parking Spaces can be achieved, and the effective area can be flexibly adapted and the algorithm error recognition rate can be reduced to a certain extent. The effective information of vehicles and owners in the work park is used to reasonably allocate parking areas for incoming vehicles, so as to reduce the cost of manual parking command and improve parking efficiency.

Problems in parking can be solved by assigning seats, and congestion can be avoided by combining route planning. If there is an opportunity later, I will consider joining the project planning. At the same time, we hope to expand the management of above-ground parking Spaces in the future, and further reduce the cost of project implementation. The detection cost of the dilution algorithm and the single-path camera can effectively monitor more parking Spaces, so as to improve the project income.