可视动目标逐步求精分割技术研究

运动目标识别跟踪系统中 ,目标分割是至关重要的处理环节。由于目标阴影、碎块等因素 ,估计精度受到影响 ,甚至导致跟踪的失败。提出了目标形状波动度的概念 ,并将其作为目标识别特征 ,利用最大似然估计准则 ,对目标初始区进行逐步求精估计。该算法简练实用 ,能有效地提高估计精度 ,为确保目标可靠性跟踪提供了新的手段。

Stepwise Accurate Segmentation for Visual Moving Targets

Targets segmentation is a fundamental part in targets tracking and recognition systems. Because of occlusions and segmentation fragments, the estimation precision is impaired, even result in an unsuccessful tracking. A new target feature named shape-fluctuation is proposed and acted as feature-discriminator in the paper. Depending on shape-fluctuation, maximum likelihood criteria and stepwise accurate estimations, optimal estimations are obtained at last. The algorithm is concise and practicable, effective in its precision-improving function, and it provides new means of guaranteeing estimation tenaciously.