A study on image sensors integrating a ultra-low-power motion vector estimator for moving picture compression

本文研究了一种基于焦平面和高速图像传感器的运动矢量估计新方法，以实现宽带移动通信中所需的低功耗运动图像编码器。在运动图像压缩中，运动矢量估计占据了很大的计算量，消耗了大量的电能。本课题负责人提出了一种利用高速中间图像估计焦平面运动矢量的新方法。这种方法称为迭代块匹配。与传统的全搜索算法相比，该算法的计算复杂度可降低1/10 ~ 1/30。一种以无损方式捕获高速中间图像的图像传感器已由首席研究员小组和松下工业公司联合研究。采用0.35μm CMOS工艺。提出了一种无损高速图像传感技术——双向多电荷转移技术。该传感器以480帧/秒的速度捕获高速图像，并以30帧/秒的速度捕获完全累积的视频图像。固定图案噪声降低到1.2mV。利用图像滞后抵消技术获得了高线性度。提出了一种基于自适应搜索区域的运动矢量估计方法。对运动图像的仿真结果表明，该方法可以有效地进一步降低算法的复杂度和降低对帧率的要求。集成运动矢量估计引擎的CMOS智能图像传感器的开发是一个近期的课题。

In this Study, a new method for motion vector estimation on focal plane and high-speed image sensors for the realization of low-power moving picture encoders which are demanded in wide-band mobile communication. The motion vector estimation dominates the computational power of the moving picture compression and consumes a large electrical power. The head researcher of this project presented a new method of motion vector estimation on focal plane using high-speed intermediate pictures. The method is called an iterative block matchins. The computational complexity can be reduced by a factor of 1/10 to 1/30 compared with the conventional full search algorithm. An image sensor which captures high-speed intermediate pictures in non-destructive manner has been implelnented as a joint-research between the group of the head investigator and Matsushita Industrial Corporation. The technology used is 0.35μm CMOS.A non-destructive high-speed image sensing technique so-called a bi-directional multiple charge transfer has been developed. The sensor captures high-speed pictures at 480frames/s and a fully-accumulated video picture at 30 frames/s. The fixed pattern noise is reduced to be 1.2mV.A high-linearity is obtained using image lag cancellation technique. A new method of motion vector estimation employing adaptive search area has also been proposed. The simulation results using moving pictures show that the proposed method is effective to further reduce the complexity and the relaxation of frame rate requirements. The development of a CMOS smart image sensor integrating the motion vector estimation engine is left as a near future subject.