基于均值偏移快速算法的红外目标跟踪

在光电跟踪设备中,传统的嵌入式跟踪器一般采用形心、相关等算法,在复杂背景下或目标受到遮挡时会丢失目标。为了能够使目标跟踪器具有抗遮挡的能力,在嵌入式目标跟踪平台上引入了均值偏移算法。在硬件的设计上,利用FPGA并行运算效率高的特点,设计了基于FPGA的直方图统计计算模块,该模块实时的将每一帧的直方图计算结果存储在SDRAM上,然后利用DSP进行均值偏移的迭代运算,在算法上针对红外图像设计了融合图像位置和像素灰度特征的改进核函数直方图作为目标特征,并提出了改进的快速均值偏移算法使其满足DSP的运算速度要求。实验表明系统在背景复杂和目标受局部遮挡时可以连续跟踪目标。跟踪性能好于传统的以型心为算法的跟踪器。在目标区域大小为64×64像素大小时,平均计算速度为22 ms。该系统和所使用的算法可支持大部分红外相机对目标的实时稳定跟踪。

Design and implementation of infrared target tracking device based on Mean Shift algorithm

In opt-electronic tracking equipment,traditional tracking device with embedded system usually adopts gravity algorithm and correlation algorithm to track the target.When the background is complex or the target is shielded partly,the device can not successfully track the target.In order to make the target tracking device have anti-occlusion ability,the mean shift algorithm is introduced on the embedded target tracking platform.In the hardware design,the characteristic that FPGA has high parallel calculation efficiency is used to design the module that performs statistic histogram calculation.The module stores the calculation result of the statistic histogram for every frame in the SDRAM in real-time.After that,a DSP is used to carry out the iterative operation of the mean shift.Aiming at the infrared video,an improved core function histogram is designed as the characteristic of the target,which combines the image position information and the grey characteristic.Finally an improved fast Mean Shift algorithm is put forward to satisfy the requirement of DSP calculation speed.Experimental results show that the proposed system can continuously track the target under complicated background and partial target occlusion conditions.The tracking performance is better than that of traditional system adopting the gravity algorithm.For the target area of 64×64 pixels,the average calculation time is 22 milliseconds per frame.The system can support most of the infrared cameras and meet the requirement of target real-time stable tracking.