PLT: Hottest temperature masking and filtering

README.md

@@ -53,6 +53,14 @@ $ ./opencv.py
 
 <br>
 
+Changes in opencv.py file:
+- Computed temperature for each pixel based on range between min/max of camera
+- Based on a temperature threshold, masked the area of interest (in this case hottest)
+- Drawing a contour of the masked area on the camera frame to mark hottest area
+- Color clustering the mask to get a mean value of the pixels in order to differentiate between regions of temperature. Each cluster's center is computed as a mean and marked by a green dot
+
+When running the program a window for each will be opened. The cluster number and threshold can be changed in the code in the call of temp_clustering method and hot_area repectively
+
 ## Related projects
 
 - https://gitlab.com/netman69/inficam

opencv.py

@@ -13,6 +13,9 @@
 cap = ht301_hacklib.T2SPLUS()
 window_name = str(type(cap).__name__)
 cv2.namedWindow(window_name, cv2.WINDOW_NORMAL)
+cv2.namedWindow("hottest_area_mask", cv2.WINDOW_NORMAL)
+cv2.namedWindow("clustered_temp", cv2.WINDOW_NORMAL)
+cv2.namedWindow("Hot_area_Contour", cv2.WINDOW_NORMAL)
 
 orientation = 0  # 0, 90, 180, 270
 
@@ -107,9 +110,42 @@ def get_fps(self):
 
     frame = rotate_frame(frame, orientation)
 
+    # Mapping temperature to pixels
+    min_temp = info['Tmin_C']
+    max_temp = info['Tmax_C']
+    temperature_image = utils.temp_img(frame, min_temp, max_temp)
+
+    # Hottest area segmentation
+    mask, hottest_area = utils.hot_area(frame, temperature_image, threshold = 28)
+    cv2.imshow('hottest_area_mask', hottest_area)
+
+    # Contouring the hottest area
+    cont_frame = utils.contoured_hottest_area(frame, mask)
+    cv2.imshow('Hot_area_Contour', cont_frame)
+
+    # Color clustering using K-means
+    clusters, cluster_centers, clustered_temp = utils.temp_clustering(hottest_area, cluster_no = 7)
+
+    # Finding cluster centroid location
+    centroid_locations = utils.cluster_centroids(cluster_centers, clustered_temp)
+    clustered_temp[centroid_locations[:, 0], centroid_locations[:, 1]] = (0, 255, 0)
+    t = '%.2fC' % max_temp
+    clustered_temp = cv2.putText(
+        clustered_temp,
+        '(Exp) PLT: ' + t,
+        (100, 185),
+        cv2.FONT_HERSHEY_PLAIN,
+        1,
+        (0, 255, 0),
+        1,
+        cv2.LINE_AA
+    )
+    cv2.imshow('clustered_temp', clustered_temp)
+
     frame = np.kron(frame, np.ones((upscale_factor, upscale_factor, 1))).astype(
         np.uint8
     )
+
     if draw_temp:
         utils.drawTemperature(
             frame,

utils.py

@@ -163,3 +163,52 @@ def get_pos(self, name, annotation_frame, roi):
         else:
             pos = name
         return pos
+
+def pixel_to_temp(pixel_value, min_temp, max_temp):
+    return ((pixel_value / 255) * (max_temp - min_temp)) + min_temp
+
+
+def temp_img(frame, min_temp, max_temp):
+    normalized_image = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
+    normalized_image = cv2.normalize(normalized_image, None, 0, 255, cv2.NORM_MINMAX)
+    return pixel_to_temp(normalized_image, min_temp, max_temp)
+
+
+def hot_area(frame, temperature_image, threshold):
+    gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
+    mask = np.where(temperature_image < threshold, 0, gray)
+    return mask, cv2.bitwise_and(frame, frame, mask=mask)
+
+
+def contoured_hottest_area(frame, mask):
+    cont_frame = frame.copy()
+    contours, hierarchy = cv2.findContours(mask, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_NONE)
+    cv2.drawContours(cont_frame, contours, -1, (0, 255, 0), 2)
+    return cont_frame
+
+
+def temp_clustering(hottest_area, cluster_no):
+    pic = cv2.cvtColor(hottest_area, cv2.COLOR_BGR2RGB)
+    pixel_vals = pic.reshape((-1, 3))
+    pixel_vals = np.float32(pixel_vals)
+
+    criteria = (cv2.TERM_CRITERIA_EPS + cv2.TERM_CRITERIA_MAX_ITER, 100, 0.9)
+    k = cluster_no
+    retval, labels, centers = cv2.kmeans(pixel_vals, k, None, criteria, 20, cv2.KMEANS_PP_CENTERS)
+
+    centers = np.uint8(centers)
+    segmented_data = centers[labels.flatten()]
+    segmented_image = segmented_data.reshape(pic.shape)
+    return k, centers, cv2.cvtColor(segmented_image, cv2.COLOR_RGB2BGR)
+
+
+def cluster_centroids(cluster_centers, clustered_temp):
+    centroid_locations = []
+
+    for row in cluster_centers:
+        cluster_i = np.where(clustered_temp == row)
+        center_x = np.mean(cluster_i[0]).astype(int)
+        center_y = np.mean(cluster_i[1]).astype(int)
+        centroid_locations.append([center_x, center_y])
+
+    return np.array(centroid_locations)