A Study Of Key Technologies Of Smart Camera Based On Reconfigurable Computing

Smart Camera is one of the research hotspots in machine vision. It has capabilities of image acquisition, processing, transmission, and advantages of modularization, miniaturization, distributed structure, ease to use, etc. It has increasing applications in object identification and location, dimension measurement, defect detection of advanced manufacturing. In this article, a smart camera system based on reconfigurable computing is proposed. The key technologies such as reconfigurable image processing pipeline architecture and FPGA implementation, collaborative design for heterogeneous computing platforms, image processing IP core design and image acquisition front-end circuit design are discussed in detail. The main contributions are as follows:An intelligent camera based on ZYNQ with reconfigurable image processing architecture is introduced, and the reconfigurable image preprocessing pipeline structure which is connected with the image acquisition module is provided. Most representative image processing algorithm IP cores such as pixel operation, filtering, edge detection, morphological processing are developed and a basic IP library is funded. The image preprocessing pipeline is tested and the reconfiguration time is about 2ms. The time consumption in FPGA is compared with that of the ARM and PC platform for different image processing algorithms. The results show that the preprocessing pipeline technology can significantly improve the image processing speed.Feature detection and matching is the most important function of smart camera. For its complexity and computing intensiveness, the computation task is split according to the specific characteristics of the algorithm, and image processing tasks are handled with the cooperation of both the hardware accelerator and the ARM processor. The collaborative design method and reconfigurable computing for ARM-FPGA heterogeneous computing platform are realized. Experiments show that the reconstruction time is at the millisecond level.Based on the in-depth analysis of CCD driving sequence, the driver for a 2 million pixel CCD image sensor with interface circuit and also the control IP core are developed to control the image acquisition front end. The performance of image acquisition reaches the design requirements.Experimental results indicate that the proposed image processing architecture can significantly improve the speed of image processing and meet the real-time requirements of smart camera.