File size: 8,502 Bytes
2cc512d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
222e5a4
2cc512d
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
14a7eac
222e5a4
14a7eac
2cc512d
14a7eac
 
 
 
 
2cc512d
14a7eac
 
 
2cc512d
14a7eac
 
 
2cc512d
14a7eac
 
 
 
 
 
 
 
 
 
222e5a4
 
14a7eac
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
222e5a4
 
14a7eac
 
 
2cc512d
 
222e5a4
2cc512d
14a7eac
222e5a4
14a7eac
 
 
 
 
 
2cc512d
 
222e5a4
 
 
 
 
 
 
 
 
 
 
 
2cc512d
 
 
 
 
 
 
 
222e5a4
2cc512d
 
 
 
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
import numpy as np
import cv2
import tensorflow as tf
from tensorflow.keras.applications import ResNet50
from tensorflow.keras.applications.resnet50 import preprocess_input
from sklearn.metrics.pairwise import cosine_similarity
from filterpy.kalman import KalmanFilter
import gradio as gr

# Load the frozen inference graph
frozen_graph_path = "frozen_inference_graph.pb"

# Load the frozen TensorFlow model
with tf.io.gfile.GFile(frozen_graph_path, "rb") as f:
    graph_def = tf.compat.v1.GraphDef()
    graph_def.ParseFromString(f.read())

# Convert the frozen graph to a function
def wrap_frozen_graph(graph_def, inputs, outputs):
    def _imports_graph_def():
        tf.compat.v1.import_graph_def(graph_def, name="")
    wrapped_import = tf.compat.v1.wrap_function(_imports_graph_def, [])
    return wrapped_import.prune(
        tf.nest.map_structure(wrapped_import.graph.as_graph_element, inputs),
        tf.nest.map_structure(wrapped_import.graph.as_graph_element, outputs))

# Define input and output tensors
inputs = ["image_tensor:0"]
outputs = ["detection_boxes:0", "detection_scores:0", "detection_classes:0", "num_detections:0"]

# Get the detection function
detection_fn = wrap_frozen_graph(graph_def, inputs, outputs)

# TensorFlow function for detection
@tf.function(input_signature=[tf.TensorSpec(shape=[1, None, None, 3], dtype=tf.uint8)])
def detect_objects(image):
    return detection_fn(image)

# Load ResNet50 for feature extraction
resnet_model = ResNet50(weights="imagenet", include_top=False, pooling="avg")

# Initialize variables to store features and identities
person_features = []
person_identities = []
person_colors = {}
kalman_filters = {}
next_person_id = 1  # Starting unique ID for persons

# Function to generate unique colors based on person ID
def get_color(person_id):
    np.random.seed(person_id)  # Ensure color is unique for each person_id
    color = tuple(np.random.randint(0, 256, size=3))  # Generates RGB tuple
    return (int(color[0]), int(color[1]), int(color[2]))  # Ensure the color is a tuple of ints

def extract_features(person_roi):
    # Resize and preprocess the ROI for ResNet50 input
    person_roi_resized = cv2.resize(person_roi, (224, 224))
    person_roi_preprocessed = preprocess_input(person_roi_resized)

    # Add batch dimension for ResNet50 input
    input_tensor = np.expand_dims(person_roi_preprocessed, axis=0)

    # Extract features using ResNet50
    features = resnet_model.predict(input_tensor)
    return features

def initialize_kalman_filter(bbox):
    kf = KalmanFilter(dim_x=7, dim_z=4)
    kf.F = np.array([[1, 0, 0, 0, 1, 0, 0],
                     [0, 1, 0, 0, 0, 1, 0],
                     [0, 0, 1, 0, 0, 0, 1],
                     [0, 0, 0, 1, 0, 0, 0],
                     [0, 0, 0, 0, 1, 0, 0],
                     [0, 0, 0, 0, 0, 1, 0],
                     [0, 0, 0, 0, 0, 0, 1]])
    kf.H = np.array([[1, 0, 0, 0, 0, 0, 0],
                     [0, 1, 0, 0, 0, 0, 0],
                     [0, 0, 0, 1, 0, 0, 0],
                     [0, 0, 0, 0, 0, 1, 0]])
    kf.R[2:, 2:] *= 10.
    kf.P[4:, 4:] *= 1000.
    kf.P *= 10.
    kf.Q[-1, -1] *= 0.01
    kf.Q[4:, 4:] *= 0.01
    kf.x[:4] = bbox.reshape((4, 1))
    return kf

def predict_bbox(kf):
    kf.predict()
    return kf.x[:4].reshape((4,))

def update_kalman_filter(kf, bbox):
    kf.update(bbox.reshape((4, 1)))
    return kf

def match_and_identify(features, bbox):
    global next_person_id

    # Flag to check if a match is found
    matched = False

    # Iterate over existing identities to check for matches
    for idx, (feat, identity) in enumerate(zip(person_features, person_identities)):
        # Compute cosine similarity between features
        similarity = cosine_similarity(
            np.array(feat).reshape(1, -1),
            np.array(features).reshape(1, -1)
        )[0][0]

        # If similarity is above threshold, consider them as the same person
        similarity_threshold = 0.7  # Adjust as needed
        if similarity > similarity_threshold:
            # Assign color if not already assigned
            if identity in person_colors:
                color = person_colors[identity]
            else:
                color = get_color(identity)
                person_colors[identity] = color

            # Update Kalman filter
            kalman_filters[identity] = update_kalman_filter(kalman_filters[identity], bbox)

            # Set matched flag to True
            matched = True
            return identity, color

    # If no match found, add new identity
    if not matched:
        person_features.append(features)
        person_identities.append(next_person_id)
        color = get_color(next_person_id)
        person_colors[next_person_id] = color

        # Initialize Kalman filter
        kalman_filters[next_person_id] = initialize_kalman_filter(bbox)

        identity = next_person_id
        next_person_id += 1

        return identity, color

def process_image(image):
    if image is None:
        print("Input image is None")
        return None

    # Convert image to RGB if it's not
    if len(image.shape) == 2:  # Grayscale
        image = cv2.cvtColor(image, cv2.COLOR_GRAY2RGB)
    elif image.shape[2] == 4:  # RGBA
        image = cv2.cvtColor(image, cv2.COLOR_RGBA2RGB)

    # Ensure image is uint8
    if image.dtype != np.uint8:
        image = (image * 255).astype(np.uint8)

    # Prepare the image tensor
    image_np = np.array(image)
    input_tensor = np.expand_dims(image_np, axis=0)

    try:
        # Run inference
        detections = detect_objects(input_tensor)

        # Extract output tensors and convert to numpy arrays
        boxes = detections[0].numpy()[0]
        scores = detections[1].numpy()[0]
        classes = detections[2].numpy()[0]
        num_detections = int(detections[3].numpy()[0])

        print(f"Number of detections: {num_detections}")

        # Filter detections for 'person' class
        threshold = 0.3  # Adjust this threshold as needed
        for i in range(num_detections):
            class_id = int(classes[i])
            score = scores[i]
            box = boxes[i]

            if class_id == 1 and score > threshold:
                h, w, _ = image.shape
                ymin, xmin, ymax, xmax = box
                left, right, top, bottom = int(xmin * w), int(xmax * w), int(ymin * h), int(ymax * h)
                
                # Extract person ROI
                person_roi = image[top:bottom, left:right]

                # Extract features
                features = extract_features(person_roi)

                # Predict bbox using Kalman filter
                predicted_bbox = np.array([xmin, ymin, xmax, ymax])

                # Match and identify
                identity, color = match_and_identify(features, predicted_bbox)

                # Draw bounding box
                left, top, right, bottom = int(predicted_bbox[0] * w), int(predicted_bbox[1] * h), int(predicted_bbox[2] * w), int(predicted_bbox[3] * h)
                cv2.rectangle(image, (left, top), (right, bottom), color, 2)
                cv2.putText(image, f'Person {identity}', (left, top - 10), cv2.FONT_HERSHEY_SIMPLEX, 0.5, color, 2)

                print(f"Detected person {identity} at ({left}, {top}, {right}, {bottom})")

    except Exception as e:
        print(f"Error during processing: {str(e)}")
        return image  # Return original image if there's an error

    return image

def gradio_interface(input_image):
    if input_image is None:
        print("Input image is None")
        return None
    
    # Convert PIL Image to numpy array if necessary
    if hasattr(input_image, 'convert'):
        input_image = np.array(input_image.convert('RGB'))
    
    # Process the input image
    output_image = process_image(input_image)
    
    if output_image is None:
        print("Output image is None")
        return None
    
    print(f"Output image shape: {output_image.shape}")
    print(f"Output image dtype: {output_image.dtype}")
    
    # Ensure the output is in the correct format for Gradio
    if output_image.dtype != np.uint8:
        output_image = (output_image * 255).astype(np.uint8)
    
    return output_image

# Create Gradio interface
iface = gr.Interface(
    fn=gradio_interface,
    inputs=gr.Image(),
    outputs=gr.Image(),
    title="Person Detection and Tracking",
    description="Upload an image to detect and track persons.",
)

# Launch the interface
iface.launch()