Attention selection model and the prior knowledge can be used in target detection and tracking effectively. This paper introduces PCNN (Pulse Coupled Neural Network) hole-filter and the prior knowledge of soccer's color based on attention selection and neural network. On one hand, in some cases the soccer is not clear, PCNN hole-filler can be used to detect its connectivity; Oh the other hand, introducing the prior knowledge of football color into the attention selection mode improves the detection accuracy. In the proposed algorithm, preprocessing techniques are utilized to remove the regions outside the filed. Then, PCNN hole-filter processes quaternion saliency map based on color differences, brightness and the prior knowledge. It can detect the soccer directly in most cases. Moreover, physical characteristics and the Kalman filter are adopted as supplementary measures to solve the cases where the football cannot be detected by PCNN directly. The computer simulation shows that compared with Dynamic Kalman Filter with Velocity Control[4] and Real Time Ball Detection Framework[5], the identification rate of our method improves 11.5% and 15.8% respectively.%注意力选择和先验知识可有效的用于目标检测与跟踪.在基于注意力选择目标跟踪模型的基础上,引入了PCNN(Pulse Coupled Neural Network)空洞滤波及足球颜色的先验知识.一方面,针对有些情况下足球模糊不清,采用PCNN空洞滤波检测足球的连通性；另一方面,在注意力选择模型中引入了足球颜色的先验知识,进一步提高检测性能.首先提取球场区域,然后对由足球颜色先验知识、色差和亮度产生的四元数显著图进行PCNN空洞滤波,很多情况下可由此直接检测到足球.如果至此未检测到足球,继续利用四元数显著图生成感兴趣区域,并用足球的面积、圆形度和离,心率等特征进一步检测目标,同时采用卡尔曼滤波嚣预测足球的位置作为补充检测.仿真结果显示,与Dynamic Kalman Filter with VelocityContro[4]和Real Time Ball Detection Framework[5]两种方法相比,检测成功率分别提高了11.5％和15.8％.
展开▼
