Object Detection Algorithm Research Based On Multiple Information Accumulation Model

Object detection as the research basis of moving object tracking, scene classification and other applications has received great attention by the computer vision expert scholars. But how to realize the fast and accurate object location under the complex background is still one of the difficulties in the object detection technology. Based on the research of the latest domestic and foreign documents, the article researches the multiple information accumulation model for the object detection as following:Firstly, considering that the sliding window detection consumes much time to search objects by using exhaustive method, we propose a fast object detection algorithm with non-sliding window based on local response accumulation combinate HOG. Fisrt of all building the components set which including the detection object’s multi-scale overlapping local areas and position relationships. Then locate the coarse positioning of the object by using the response results and position constraint of the projection detectors, by using this method much position of non-object needn’t to research. In order to locate the object accurately, we use the feature of HOG and the SVM classifier to validate the object detection’s result.Secondly, considering the based on cascade classification model for the human eye detection algorithm easily mistakes the human eyes to the eyebrows area the problem, we put forward the eye accurate location algorithm of the color information aiding the cascade vote. Firstly we select the feature of Haar and Ada Boost cascade classifier to locate the coarse position of the face. Then use the ballot results of the organs’ number in the face region to exclude false face and locate the eyes at the same time. Next we locate the eyes by using the skin color model, maximum between-cluster variance and optimization of integral projection on the upper face area. In order to locate precise position of the eyes, we use the skin color model aiding cascade eye location to correct error eyes’ position. The experimental results show that this algorithm has obvious advantage of locating eyes on all kinds of complicated situationsFinally, considering that it is difficult to locate the license plate in the open environment, we propose a license plate localization algorithm based on the local character set and prior knowledge. Firstly building local character set by using a small amount of representative local features and the position relationships of the license plate character. Then we selecte feature points of the license plate using the matching results of the local character set and the tested image. Next eliminate false matching plates feature points basing on the license plate prior knowledge. According to the relative position of the license plate characters in the detection image and the local character set we can locate the plate. This algorithm is robust to locate the license plate under the open environment and can realize the real-time and accurate license plate detection.