Bus Passenger Counting Video Stream Preprocessing System Accelerated By FPGA

Bus passenger traffic statistics are of great significance for building a smart bus system.The automatic passenger counting method based on visual monitoring technology provides a solution for rational dispatching of bus resources.However,the large amount of image and video data makes it difficult for many image algorithms running on ordinary serial processors to meet the real-time requirements in practical application scenarios.Programmable Logic Array FPGAs have a wealth of hardware logic resources,and their high-speed parallel computing capabilities and pipeline structure design advantages make it suitable for parallelization of most image algorithms.In this paper,FPGA is used as the core processor,and the parallelization processing and pipeline structure design of image preprocessing part in human head detection and counting are studied.The main contents are as follows:1.Built a series of hardware system framework including CMOS camera image acquisition,SDRAM image storage,FPGA image parallel processing and VGA image display.According to the timing and function requirements of CMOS camera,SDRAM memory and VGA protocol,image data acquisition of CMOS camera,read and write control of SDRAM memory and normal display under VGA protocol are realized.2.The calculation of the proportion of the calculation amount and the possibility of parallelization design of each algorithm in the head detection counting method are analyzed,and the algorithm which contains a large number of multiplication and addition operations and is suitable for parallelization is selected for acceleration.Accelerated processing of color image grayscale,Gaussian smoothing,Canny edge detection and background erasure is realized by pipeline and parallel structure.3.The experimental waveform and the resulting image show that under the same conditions,the human head detection preprocessing using FPGA parallel acceleration can accurately remove the complex background noise and extract the edge contour of the moving target,which can significantly improve the video compared with the PC serial processing.The image preprocessing speed can be up to 26 times.