Segmentation的实现好像与官方不一样。

[mozheng]

问个问题，官方demo的Segmentation不是你这样写的。你这样写是什么意思？实验证明这么写的确会减缓启动时噪点多的问题，但也会失去部分精度。

def Segmentation(self, uint8_t[:, :] image):        
        cdef int width = image.shape[1]
        cdef int height = image.shape[0]
        
        # cdef np.ndarray[uint8_t, ndim=2] segmentation = np.zeros([width, height], dtype=np.uint8)
        cdef uint8_t[:, :] segmentation_map = np.empty((height, width), dtype=np.uint8)  # 官方的 segmentation_map每次调用不是申请为0的区域，而是复用上一次的segmentation_map
        
        cdef uint8_t *image_data_ptr = &image[0, 0]
        cdef uint8_t *segmentation_map_ptr = &segmentation_map[0, 0]
        
        # call C function
        cdef int32_t back
        back = c_vibe.libvibeModel_Sequential_Segmentation_8u_C1R(self.model, image_data_ptr, segmentation_map_ptr)
        # Error: need convert segmentation_map to numpy.array
        # return segmentation_map
        return np.asarray(segmentation_map)

下面是官方的实现

while ((char)keyboard != 'q' && (char)keyboard != 27) {
    /* Read the current frame. */
    if (!capture.read(frame)) {
      cerr << "Unable to read next frame." << endl;
      cerr << "Exiting..." << endl;
      exit(EXIT_FAILURE);
    }

    if ((frameNumber % 100) == 0) { cout << "Frame number = " << frameNumber << endl; }

    if (frameNumber == 1) {
      segmentationMap = Mat(frame.rows, frame.cols, CV_8UC1);
      model = (vibeModel_Sequential_t*)libvibeModel_Sequential_New();
      libvibeModel_Sequential_AllocInit_8u_C3R(model, frame.data, frame.cols, frame.rows);
    }

    libvibeModel_Sequential_Segmentation_8u_C3R(model, frame.data, segmentationMap.data);
    libvibeModel_Sequential_Update_8u_C3R(model, frame.data, segmentationMap.data);
    medianBlur(segmentationMap, segmentationMap, 3); /* 3x3 median filtering */

  }

[232525]

你看一下这一行/~https://github.com/232525/ViBe.Cython/blob/master/lib_vibe/vibe-background-sequential.c#L321 ，官方实现应该也不是复用上一次结果，好像是把数据清成1，segmentation_map是每一次调用模型获取前景分割的结果，不至于需要复用的，我改的时候只是单纯的为了可以python调用这个函数，没考虑太多。