Vehicle Detection in Open Parks Using a Convolutional Neural Network

This paper proposed a new vehicle detection algorithm based on a CNN (convolutional neural network), which dedicates to detect and localize vehicles in an open park. After an off-line training the network can fast respond to an input image so that it is suitable for real-time applications and has the potential to use in vehicle park management systems. Firstly, the trained CNN with a defined sliding window is used to search and identify vehicles in open parks. Secondly, a distribution matrix is defined to reflect the density of vehicle distribution, and it is used to remove redundant windows of vehicles to locate a position of vehicle accurately. Compared to other approaches for vehicle detection, the CNN-based approach does not require any engineered features. The proposed algorithm has combined a CNN with the distribution matrix so that the accuracy of the position location has been improved.