复杂物体形状特征及孔洞特征的高速提取

在以机器人视觉及自动驾驶为代表的实时图像识别系统中，尽快计算图像中物体的相关特征是进行物体识别的不可或缺的先决条件。提取与物体相关的特征越多越快，系统就越能对物体进行精准识别。本项目将从三个方面开展研究：首次定义复杂物体中与孔洞相关的特征并提出同时计算物体形状特征和孔洞特征的算法；彻底颠覆依赖标记处理所生成的标记图像进行物体形状特征计算的现有方法，提出不需要生成标记图像就可以同时计算复杂物体形状特征和孔洞特征的崭新方法，将需要扫描图像的次数从3次降为1次，从而大幅度减少计算物体形状特征和孔洞特征的时间和存储空间；基于上述1次扫描算法，提出适合流水线技术和并行处理的相应算法并用FPGA加以实现，为设计制造提取物体形状特征和孔洞特征的专用芯片打下基础。本项目的研究成果将增加用于识别物体的特征要素、缩短提取复杂物体形状特征和孔洞特征的时间、提高实时图像识别系统的整体性能、推动计算机视觉的发展。

In real-time image processing systems such as robot vision and auto-driving systems, extracting objects’ relative features as fast as possible is indispensable for object recognition. More and faster for extracting objects’ relative features, more accurate for object recognition. This project will do relative researches in three ways: (1) define features relative to holes in complex objects for the first time, and propose an algorithm for calculating objects’ shape features and holes’ features simultaneously; (2) completely change the method for calculating objects’ features by use of the labeled image generated by labeling processing, propose an absolutely new algorithm for calculating objects’ shape features and holes’ features simultaneously without use of the labeled image, in which the necessary times for scanning a giving image will decrease from three to one, thus, it can reduce the time and memory space for calculating the objects’ shape features and holes’ features; (3) based on the above proposed one-scan algorithm, propose an algorithm suitable for pipeline processing and parallel processing, and implement the algorithm by use of FPGA, which will be a base for designing and manufacturing the chip special for extracting objects’ shape features and holes’ features. With the achievements of this project, the factors used for object recognition will increase a lot; the time for calculating objects’ shape features and holes’ features will decrease greatly; the performance of real-time image recognition systems can be improved. In this way, it can push forward the development of computer vision.