基于联合双边滤波器的Kinect深度图像滤波算法

Kinect depth image filtering algorithm based on joint bilateral filter

针对Kinect镜头采集的深度图像一般有噪声和黑洞现象,直接应用于人体动作跟踪和识别等系统中效果差的问题,提出一种基于联合双边滤波器的深度图像滤波算法。算法利用联合双边滤波原理,将Kinect镜头同一时刻采集的深度图像和彩色图像作为输入,首先,用高斯核函数计算出深度图像的空间距离权值和RGB彩色图像的灰度权值;然后,将这两个权值相乘得到联合滤波权值,并利用快速高斯变换替换高斯核函数,设计出联合双边滤波器;最后,用此滤波器的滤波结果与噪声图像进行卷积运算实现Kinect深度图像滤波。实验结果表明,所提算法应用在人体动作识别和跟踪系统后,可显著提高在背景复杂场景中的抗噪能力,识别正确率提高17.3%,同时所提算法的平均耗时为371 ms,远低于同类算法。所提算法保持了联合双边滤波平滑保边的优点,由于引入彩色图像作为引导图像,去噪的同时也能对黑洞进行修补,因此该算法在Kinect深度图像上的去噪和修复效果优于经典的双边滤波算法和联合双边滤波算法,且实时性强。

Usually the depth image obtained by Kinect camera contains noise and black holes, so the effect is poor if it is directly applied into human motion tracking and recognition system. To solve this problem, an efficient depth image filtering algorithm based on joint bilateral filter was proposed. The principle of joint bilateral filtering was used in the proposed algorithm, and the depth and color images were captured by Kinect camera at the same time as the input. Spatial distance weight value of depth image and grayscale weight value of RGB color image were computed by Gaussian kernel function. Then these two weight values were multiplied to get the weight value of joint bilateral filter. A joint bilateral filter was designed by replacing the Gaussian kernel function with fast Gaussian transform. Finally, this filtered result was convolved with the noisy image to filter the Kinect depth image. The experimental results show that the proposed algorithm can significantly improve the robustness to noise in the human motion tracking and identification system and increase the recognition rate by 17.3%. The average running time of the proposed algorithm is 371ms, and is much lower than similar other algorithms. The proposed algorithm keeps the advantages of joint bilateral filter. Since the color image is introduced into the algorithm, the proposed algorithm can well repair the black holes while reducing the noise. The proposed algorithm is better than traditional bilateral filter and joint bilateral filter in denoising and repairing holes for the Kinect depth image, and it has higher real-time performance.