File size: 10,104 Bytes
14e0055 f9ccd6b 14e0055 f9ccd6b 14e0055 6fa13e0 14e0055 6fa13e0 14e0055 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 |
import numpy as np
from scipy.ndimage import convolve
import os
from numpy.lib.stride_tricks import sliding_window_view
from numba import njit
import cv2
def mipi_raw10_to_raw8_scaled(raw10_data):
raw10_data = np.frombuffer(raw10_data, dtype=np.uint8)
n_blocks = len(raw10_data) // 5
raw10_data = raw10_data[:n_blocks * 5].reshape(-1, 5)
p0 = (raw10_data[:, 0].astype(np.uint16) << 2) | ((raw10_data[:, 4] >> 0) & 0x03)
p1 = (raw10_data[:, 1].astype(np.uint16) << 2) | ((raw10_data[:, 4] >> 2) & 0x03)
p2 = (raw10_data[:, 2].astype(np.uint16) << 2) | ((raw10_data[:, 4] >> 4) & 0x03)
p3 = (raw10_data[:, 3].astype(np.uint16) << 2) | ((raw10_data[:, 4] >> 6) & 0x03)
raw8_data = np.empty((n_blocks * 4 * 2,), dtype=np.uint8)
raw8_data[0::8] = p0 & 0xFF
raw8_data[1::8] = p0 >> 8
raw8_data[2::8] = p1 & 0xFF
raw8_data[3::8] = p1 >> 8
raw8_data[4::8] = p2 & 0xFF
raw8_data[5::8] = p2 >> 8
raw8_data[6::8] = p3 & 0xFF
raw8_data[7::8] = p3 >> 8
return raw8_data.tobytes()
def readRAW(path):
filesize = os.path.getsize(path)
with open(path, "rb") as f:
raw_data = f.read()
# Case 1: 如果是 MIPI RAW10 格式,大小为 7,372,800 字节
if filesize == 7372800:
raw_data = mipi_raw10_to_raw8_scaled(raw_data)
# 转换为 int16 并 reshape
arr = np.frombuffer(raw_data, dtype=np.int16).reshape(96, 240, 256)
# Byte Swap: [x,y,256] → [x,y,128,2] → swap last dim → [x,y,256]
reshaped = arr.reshape(*arr.shape[:-1], -1, 2)
swapped = reshaped[..., ::-1]
histogram_data = swapped.reshape(arr.shape)
# Line remapping (每组8行:0,4,1,5,...)
mapping = [0, 4, 1, 5, 2, 6, 3, 7]
group_size = 8
num_groups = 12 # 96 / 8
output = np.empty_like(histogram_data)
for g in range(num_groups):
start = g * group_size
end = start + group_size
output[start:end, :, :] = histogram_data[start:end, :, :][mapping, :, :]
return output.astype(np.int16)
def binning_2x2_stride2(data):
"""
data: numpy array (96, 240, 256)
return: numpy array (48, 120, 256)
"""
# 先 reshape 再求和,效率高
return data.reshape(48, 2, 120, 2, 256).sum(axis=(1, 3))
def binning_2x2_stride1(data):
"""
data: numpy array (96, 240, 256)
return: numpy array (95, 239, 256) # 因为stride=1,边界少1行1列
"""
# 直接用切片叠加四个偏移
binned = (data[:-1, :-1] + data[1:, :-1] +
data[:-1, 1:] + data[1:, 1:])
# binned = np.pad(binned, ((0,1),(0,1),(0,0)), mode='constant')
return binned
def ma_vectorized(data, kernel=[-2, -2, 1, 2, 2, 3, -1, -1]):
kernel = np.array(kernel, dtype=np.float32)
k = kernel.size
kernel_sum = kernel.sum()
# 确保 data 是 numpy array
data = np.asarray(data, dtype=np.float32)
# 取滑动窗口视图,shape: (96, 240, 256 - k + 1, k)
windows = np.lib.stride_tricks.sliding_window_view(data, window_shape=k, axis=2)
# 直接点乘kernel,然后求和,得到平滑结果 (96,240,256-k+1)
smoothed = np.tensordot(windows, kernel, axes=([3],[0])) / kernel_sum
smoothed[smoothed<0] = 0
# 为了保持和输入长度一致,可以两边补0或其他策略,这里简单在尾部补零
pad_width = ((0,0), (0,0), (0,k-1))
smoothed = np.pad(smoothed, pad_width, mode='constant', constant_values=0)
return smoothed
def ma_vectorized_fast(data, kernel=[-2, -2, 1, 2, 2, 3, -1, -1]):
kernel = np.array(kernel, dtype=np.float32)
k = kernel.size
kernel_sum = kernel.sum()
if kernel_sum == 0:
kernel_sum = 1 # 避免除 0
# padding 边界,保持中心对齐
pad = k // 2
data_padded = np.pad(data, ((0,0),(0,0),(pad,pad)), mode='edge')
# 利用 np.convolve 沿最后一个轴计算
def conv_1d(x):
return np.convolve(x, kernel[::-1], mode='valid') / kernel_sum
# 按最后一维应用
smoothed = np.apply_along_axis(conv_1d, 2, data_padded)
smoothed = np.maximum(smoothed, 0) # 负数置零
return smoothed
BIN_SIZE = 180
MAX_PEAKS = 2 # 峰值个数
@njit
def sum_hist(hist, length):
return np.sum(hist[:length])
@njit
def max_hist(hist, length):
return np.max(hist[:length])
@njit
def compute_centroid(hist, start_bin, end_bin):
bins = np.arange(start_bin, end_bin + 1)
values = hist[start_bin:end_bin + 1]
total = np.sum(values)
if total == 0:
return (start_bin + end_bin) / 2.0
return np.sum(bins * values) / total
@njit
def find_peaks_hw(histograms, histograms_ma):
"""
输入:
histograms: (H, W, 256) 原始直方图
histograms_ma: (H, W, 256) 平滑直方图
输出:
tof_data: (H, W, MAX_PEAKS) 质心位置
peak_data: (H, W, MAX_PEAKS) 峰值强度
noise_data: (H, W) 噪声值
multishot_data: (H, W) 多拍信息
totalcount: (H, W) 总计数
nt_count_n: (H, W, MAX_PEAKS) NT计数
"""
H, W, _ = histograms.shape
# 输出初始化
tof_data = np.zeros((H, W, MAX_PEAKS), dtype=np.float32)
peak_data = np.zeros((H, W, MAX_PEAKS), dtype=np.int32)
noise_data = np.zeros((H, W), dtype=np.int16)
multishot_data = np.zeros((H, W), dtype=np.int16)
totalcount = np.zeros((H, W), dtype=np.int32)
nt_count_n = np.zeros((H, W, MAX_PEAKS), dtype=np.int32)
for i in range(H):
for j in range(W):
hist_raw = histograms[i, j]
hist = histograms_ma[i, j]
count = 0
bin_idx = 1
# 多拍信息
multishot = (int(hist_raw[254]) << 8) + (int(hist_raw[255]) >> 2)
multishot_data[i, j] = multishot
totalcount[i, j] = int(sum_hist(hist_raw, BIN_SIZE) << 13) // multishot
# 噪声估计
noise_data[i, j] = int(sum_hist(hist, 8) + 4) >> 3
noise_i = max_hist(hist, 8)
# max_th = np.max(hist) * 0.01 + 25
max_th = 50
noise_i = max_th if noise_i < max_th else noise_i * 3
th = 2 * noise_i - 24 / (2e4 * 4) * noise_i * noise_i
# 峰值检测
while bin_idx < BIN_SIZE - 1 and count < MAX_PEAKS:
# 找到第一个 >= TH 的 bin
while bin_idx < BIN_SIZE - 1 and hist[bin_idx] < th:
bin_idx += 1
bin_idx -= 1
start_bin = bin_idx
start_peak = hist[start_bin]
# 找最大值 bin
while bin_idx + 1 < BIN_SIZE and (
hist[bin_idx] < hist[bin_idx - 1] or hist[bin_idx] < hist[bin_idx + 1]
):
bin_idx += 1
max_bin = bin_idx
max_peak = hist[max_bin]
# 找 end_bin
while bin_idx + 1 < BIN_SIZE and (
hist[bin_idx] > th or hist[bin_idx] > start_peak or hist[bin_idx] > hist[bin_idx + 1]
):
bin_idx += 1
end_bin = bin_idx
if (
start_bin == end_bin
or start_bin == max_bin
or max_bin == end_bin
or (max_peak - start_peak) < 50
):
bin_idx += 1
continue
# 质心
centroid = compute_centroid(hist, start_bin, end_bin)
tof_data[i, j, count] = centroid
peak_data[i, j, count] = (int(hist[max_bin]) << 13) // multishot
# NT count
nt_end_bin = max_bin - 10
nt_start_bin = nt_end_bin - 48
nt_start_bin = 0 if nt_start_bin < 0 else nt_start_bin
nt_num = nt_end_bin - nt_start_bin
est_nt = 48 * noise_data[i, j] if nt_num < 48 else 0
nt_count_n[i, j, count] = np.sum(hist[nt_start_bin:nt_start_bin + nt_num]) + est_nt
count += 1
peak_data=peak_data*256/48000
return tof_data, peak_data, noise_data, multishot_data, totalcount, nt_count_n
def local_threshold(img, window_size=15, C=2):
h, w = img.shape
out = np.zeros_like(img, dtype=np.uint8)
pad = window_size // 2
padded = cv2.copyMakeBorder(img, pad, pad, pad, pad, cv2.BORDER_REFLECT)
for i in range(h):
for j in range(w):
local_region = padded[i:i+window_size, j:j+window_size]
# thresh = np.mean(local_region) - C
thresh = np.mean(local_region,axis=(0,1)) - 0.1
out[i,j] = img[i,j] if img[i,j] > thresh else 0
return out
def select_peaks_hw(tof_data, peak_data):
"""
输入:
tof_data: (H, W, MAX_PEAKS) 质心位置
peak_data: (H, W, MAX_PEAKS) 峰值强度
输出:
tof: (H, W)
peak: (H, W)
"""
# peak_data = peak_data *256/48000
ref_set = np.zeros_like(peak_data)
for i in range(MAX_PEAKS):
ref_set[...,i] = peak_data[...,i]*tof_data[...,i]*tof_data[...,i] /1200 * 6
ref = np.log2(ref_set[...,i])
ref[ref<0] = 0
_, otsu_binary = cv2.threshold(ref.astype(np.uint8), 0, 255, cv2.THRESH_BINARY + cv2.THRESH_OTSU)
# dynamic_ref = np.max(ref,axis=(0,1)) * 0.8
# ref[ref>100] = 100
# ref[ref<3] = 0
# ref = local_threshold(ref,5,0)
ref_set[...,i] =(ref)
# ref_set[...,1]*= 3
ref = np.max(ref_set,axis=2)
frame1 = ref_set[...,0]
frame2 = ref_set[...,1]
# 两个 mask 初始化为 0
mask1 = np.zeros_like(frame1, dtype=np.uint8)
mask2 = np.zeros_like(frame2, dtype=np.uint8)
# 逐像素比较:frame1 > frame2
mask1[(frame1 > frame2) & (frame1 > 0)] = 1
mask2[(frame2 > frame1) & (frame2 > 0)] = 1
# 如果相等且非零,可以任选一帧,这里给 mask1
mask1[(frame1 == frame2) & (frame1 > 0)] = 1
tof = tof_data[...,0] * mask1 + tof_data[...,1] * mask2
peak = peak_data[...,0] * mask1 + peak_data[...,1] * mask2
return tof,peak,ref,ref_set |