The Vehicle License Plate Recognition Research Of Imitating Human Vision Under Conditions Of Uncertainty

Detection and identification of vehicle license plate is an important topic in the research of computer vision and image processing, and it's a typical area in target detection. Due to the change of weather and light, and other disturbance such as complex background, discrepancy is often found between the description of vehicle license plate and the actual object, which brings great challenge to detect and identify a vehicle license plate correctly. In allusion to the drawbacks of low adaptability and robustness of current visual system, this thesis is to research the mechanism and method of target detection and identification. The details are as below:Firstly, the human vision's mechanism of target detection is studied. First of all, the mechanism of human vision is studied, and on the basis of researching on visual characteristics of objects and human visual perception characteristics, the human visual perception knowledge-based inference and human visual knowledge-based and coarse-to-fine multi-level recognition mechanism are analyzed. Then, the regulation of human visual perception is summed up, as well as the humanoid vision, after which the closed-loop feedback model of visual detection system is established.Secondly, the segmentation of vehicle license plate by layered perception is studied. First of all, vehicle as a whole is detected on a large scale image, and the modeling method for background is realized based on the multi-frame evidence, as well as the algorithm on real-time updating background based on the inter-frame. Then, the vehicle license plate is checked on a small scale of the vehicle and the vehicle license plate as whole is checked by adopting the bridge method to ignore the unnecessary details on a smaller scale image. Last but not the least, the position setting of the vehicle license plate basing on a feedback mechanism is proposed, as well as the multi-scale segmentation method to the character of a vehicle license plate.Thirdly, the vehicle license plate is recognized without sufficient information. Taking a moving car with the vehicle license plate partially covered as a study object and based on the rule that human's vision can deduce from locality to full, the license plate is checked integrated with multi-frame evidences. To study a vehicle license plate with its information partially missing, the approximate area of the doubtful vehicle is checked and ratiocinated according to the multi-frame characteristics. By adopting the block idea in the recognition of characters, the vehicle license plate recognition method is proposed.It's proved by lots of experiments that locating a vehicle license plate can effectively solve problems under the circumstances such as uneven illumination, depigmentation or other complex environments. This recognition algorithm can recognize missing characters, and can be used to gradually recognize a license plate which is been partially shaded. The new way of calculation and identification is capable to improve the robustness and adaptability of the automatic recognition of a vehicle license plate.