Automatic Parking System Based On Surround View Vision

The complex and diverse parking environment require high driving level for drivers,and a little carelessness will cause the vehicle to collide with surrounding obstacles or pedestrians,causing traffic accidents.How to park quickly and safely has become a headache for many drivers.In addition,through the rapid development of autonomous driving technology,driverless cars also urgently need to solve the problem of unattended automatic parking.In order to solve this problem,we intend to study the fully automatic unmanned parking technology for the surround vision scene and propose some innovative algorithms in parking space detection and empty parking space classification.Specific research contents include:(1)An optimized parking space detection algorithm is proposed,which significantly improves the detection accuracy and speed.In terms of corner detection,the introduction of focus loss alleviates the problem of positive and negative sample imbalance in the single-step detection algorithm;the introduction of a global attention mechanism to build the dependency relationship between channels to achieve the enhancement of important features,so as to obtain performance Improvement;by introducing a deep separable convolution structure,the network calculation is greatly reduced,and the forward inference speed is significantly improved.The final algorithm has an accuracy rate and recall rate of 99.72% and 99.17% on public data sets,respectively.(2)An empty parking space classification algorithm based on perspective transformation is proposed.In order to make the parking space detection algorithm more complete,and to address the data dependency problem and the need to set fixed parameters in the current mainstream empty parking space classification algorithm,this paper proposes a generalization based on perspective transformation and deep convolutional neural network technology An empty parking space classification algorithm with strong performance and no need to set fixed parameters.Experiments on public data sets show that the proposed algorithm achieves a significant performance improvement over existing algorithms.(3)Designed and implemented an automatic parking system based on the BS architecture.In order to verify the actual effect of the algorithm proposed in this paper,we designed and implemented an automatic parking simulation system that is easy to operate and visualize the results based on the BS architecture.In order to build a complete automatic parking system,in addition to the other algorithms in this article,we have also retained path planning,results display and other related modules.