基于热成像的巡检及AidLux方案实现

def process_img(img, target_size=640, max_size=2000, multiple=32, keep_ratio=True, NCHW=True, ToTensor=True):
img = img128:512, 0:480
img = cv2.resize(img, (640, 512), interpolation=cv2.INTER_LINEAR)
im_shape = img.shape
im_size_min = np.min(im_shape0:2)
im_size_max = np.max(im_shape0:2)

if keep_ratio:
im_scale = float(target_size) / float(im_size_min)
if np.round(im_scale * im_size_max) > max_size:
im_scale = float(max_size) / float(im_size_max)
im_scale_x = np.floor(img.shape1 * im_scale / multiple) * multiple / img.shape1
im_scale_y = np.floor(img.shape0 * im_scale / multiple) * multiple / img.shape0
image_resized = cv2.resize(img, None, None, fx=im_scale_x, fy=im_scale_y, interpolation=cv2.INTER_LINEAR)
im_scales = np.array(im_scale_x, im_scale_y, im_scale_x, im_scale_y)
im = image_resized / 255.0 # np.float64
im = im.astype(np.float32)
PIXEL_MEANS =(0.485, 0.456, 0.406) # RGB format mean and variances
PIXEL_STDS = (0.229, 0.224, 0.225)
im -= np.array(PIXEL_MEANS)
im /= np.array(PIXEL_STDS)
im = cv2.cvtColor(im, cv2.COLOR_BGR2RGB) # BGR2RGB
if NCHW:
im = np.transpose(im, (2, 0, 1)).astype(np.float32) # SAI-KEY TensorFlow use input with NHWC.
im = imnp.newaxis, ...
if ToTensor:
im = torch.from_numpy(im)
return im, im_scales
else:
return None 

if __name__=="__main__":
tflite\_model = '/home/R-RetinaNet/models/r-retinanet.tflite'
# 定义输入输出shape
in\_shape = [1 \* 640 \* 800 \* 3 \* 4]  # HWC, float32
out\_shape = [1 \* 53325 \* 8 \* 4]  # 8400: total cells, 52 = 48(num\_classes) + 4(xywh), float32
# AidLite初始化
aidlite = aidlite\_gpu.aidlite()
# 加载R-RetinaNet模型
res = aidlite.ANNModel(tflite\_model, in\_shape, out\_shape, 4, -1) # Infer on -1: cpu, 0: gpu, 1: mixed, 2: dsp
# print(res)
'''
读取手机实时摄像头数据
'''
cap = cvs.VideoCapture(0)
frame\_id = 0
while True:
    frame = cap.read()
    if frame is None:
        continue
    im, im\_scales = process\_img(frame, NCHW=False, ToTensor=False)  # im: NHWC
    frame\_id += 1
    if frame\_id % 3 != 0:
        continue
    aidlite.setInput\_Float32(im, 800, 640)
    # 推理
    aidlite.invoke()
    preds = aidlite.getOutput\_Float32(0)
    preds = preds.reshape(1, 8, (int)(preds.shape[0]/8))
    # 后解算
    output = np.transpose(preds, (0, 2, 1))
    # 创建Anchor
    im\_anchor = np.transpose(im, (0, 3, 1, 2)).astype(np.float32)
    anchors\_list = []
    anchor\_generator = Anchors(ratios = np.array([0.2, 0.5, 1, 2, 5]))
    original\_anchors = anchor\_generator(im\_anchor)   # (bs, num\_all\_achors, 5)
    anchors\_list.append(original\_anchors)
    # 解算输出
    decode\_output = decoder(im\_anchor, anchors\_list[-1], output[..., 5:8], output[..., 0:5], thresh=0.2, nms\_thresh=0.2, test\_conf=None)
    # 重构输出
    scores = decode\_output[0].reshape(-1, 1)
    classes = decode\_output[1].reshape(-1, 1)
    boxes = decode\_output[2]
    boxes[:, :4] = boxes[:, :4] / im\_scales
    if boxes.shape[1] > 5:   
        boxes[:, 5:9] = boxes[:, 5:9] / im\_scales
    dets = np.concatenate([classes, scores, boxes], axis=1)
    # 过滤类别
    keep = np.where(classes > 0)[0]
    dets =  dets[keep, :]
    # 转换坐标('xywha'->'xyxyxyxy')
    res = sort\_corners(rbox\_2\_quad(dets[:, 2:]))
    # cv绘图.
    for k in range(dets.shape[0]):
        cv2.line(frame, (int(res[k, 0]), int(res[k, 1])), (int(res[k, 2]), int(res[k, 3])), (0, 255, 0), 3)
        cv2.line(frame, (int(res[k, 2]), int(res[k, 3])), (int(res[k, 4]), int(res[k, 5])), (0, 255, 0), 3)
        cv2.line(frame, (int(res[k, 4]), int(res[k, 5])), (int(res[k, 6]), int(res[k, 7])), (0, 255, 0), 3)
        cv2.line(frame, (int(res[k, 6]), int(res[k, 7])), (int(res[k, 0]), int(res[k, 1])), (0, 255, 0), 3)
    cvs.imshow(frame)