Real-time motion tracking by using spatial reticles inhomogeneously distributed according to the spatially variant distribution of receptive fields of visual peripheral nerves.

开发了一种具有实时操作的视觉跟踪系统，用于保持运动对象的图像的大小和位置。在该系统中，采用了传感器分布的空间变化方案，为分析物体运动的全局运动参数提供了有效的手段。为了实现这一点，在第一个项目年，通过模仿视觉外周神经接受野的组织，建立了一个Gabor掩模库。单个掩模作为速度传感器，在图像平面上产生速度矢量场，速度传感通常需要较宽的视场来平均图像运动，而每个掩模的有限大小会导致速度估计的降低。图像运动的Radon变换由于其投影的整体特性，给出了解决这一问题的一种方法。图像运动中包含的平移和膨胀分量可以通过评估Radon域上的光流来分离、识别和测量，将Radon变换应用到运动目标的动态场景中，使我们能够使用单目摄像系统来视觉跟踪目标。引入卡尔曼滤波对目标位置进行平滑预测，改善了延迟特性，从而减小了定位误差。该系统仅由一台商用PC机组成，具有良好的跟踪性能和全视频帧的处理速率。

A visual tracking system with real time operation for maintaining both the size and the position of the image of a moving object has been developed. In the system, a spatially variant scheme of sensor distribution is adopted for providing an effective means of analyzing global motion parameters of object motions. To achieve this, a bank of Gabor reticles was constituted in the 1st project year by mimicking the organization of the receptive field of visual peripheral nerves. The individual reticle works as a velocity sensor, and thus yielding a velocity vector field on the image plane.Velocity sensing generally requires a wide field of view for averaging image motion, and the limited size of each reticle causes degradation in velocity estimation. The Radon transform of image motion yields a solution to this problem because of its integral feature of projection. Both translation and dilation components included in image motion can be isolated, identified and measured by evaluating the optical flow on the Radon domain.Applying the Radon transform to the dynamic scene of a moving object enables us to visually track the object with a monocular camera system. A Kalman filter is also introduced to smoothly predict the position of the object and improve the delay characteristics, thereby reducing the error of localization. The proposed system is composed of only a commercially available PC, and exhibits good tracking performances with processing rate of full video frames.