Home Safety Using Fog Computing

Alexa with notifications on Hazardous event
Home Safety and Security- Web Testing Interface
Notification
False Positive
True Negative
True Positive

Inspiration

Seeing last thanksgiving havoc wrecking boulder fire, we started wondering can there be a way this could have been prevented. With the advancement in the field of Engineering and computer we felt that we should have done more to prepare for such tragic events. Starting with this objective we started working on ways we can do early detection of fires and alarm the authority to take appropriate actions at inappropriate times.

One common scenario that we observed during the news coverage of boulder fires was that there were many household with security features like CCTV cameras, anti theft alarms etc. But these systems were not capable enough to detect a natural event like fire and send SOS calls to avoid deadly consequences. There were several households who left behind a pet at home and were out for work or the thanksgiving break and were not able to save their pets on time.

Although the security camera's feed was viewing images of fire/smoke outside the, it wasn't alarming the homeowners about it.

What it does

We have developed a low-cost system that can enable this security camera to be able to do computations on the image captured to the other household devices with computing power and trigger events if there is any significant change in the stimuli.

Neural Network Model

from tensorflow.keras.optimizers import RMSprop,Adam
model = tf.keras.models.Sequential([
          tf.keras.layers.Conv2D(96, (11,11), strides=(4,4), activation='relu', input_shape=(224, 224, 3)),
          tf.keras.layers.MaxPooling2D(pool_size = (3,3), strides=(2,2)),
          tf.keras.layers.Conv2D(256, (5,5), activation='relu'),
          tf.keras.layers.MaxPooling2D(pool_size = (3,3), strides=(2,2)),
          tf.keras.layers.Conv2D(384, (5,5), activation='relu'),
          tf.keras.layers.MaxPooling2D(pool_size = (3,3), strides=(2,2)),
          tf.keras.layers.Flatten(),
          tf.keras.layers.Dropout(0.2),
          tf.keras.layers.Dense(2048, activation='relu'),
          tf.keras.layers.Dropout(0.25),
          tf.keras.layers.Dense(1024, activation='relu'),
          tf.keras.layers.Dropout(0.2),
          tf.keras.layers.Dense(2, activation='softmax')
])
model.compile(loss='categorical_crossentropy',
              optimizer=Adam(lr=0.0001),
              metrics=['acc'])
model.summary()

How we built it

In our project, we have designed a Neural network Deep Learning model to identify Fire in images with an accuracy of 98%. This trained model is then extracted as an ML as a service and used to serve API requests from different systems. Now a camera in the home network can send an image to a let's say a PC running this model and this model can predict whether there is any hazardous condition in the image. In the event of any life-threatening scenario, the model sends an SOS message to the owner using an SMS service. It is also capable of announcing a notification on an Alexa Echo Dot. We have used two ways of communication to ensure that the ALERT doesn't go overheard.

Challenges we ran into

Finding Datasets, Training with limited GPU capabilities. Integrating ALEXA into the system for alerting homeowners. Integrating Twilio for SMS

Accomplishments that we're proud of

The model is designed in a way to be lite weight and can be run on spare mobile devices as well.

It is also ensuring the privacy of the user, as it is not sending any data over to the cloud. Removing the use of Cloud Technology in the systems also makes the systems extremely cheap and no recurrent cost is attached to them.

Using TensorFlow Lite will help us address key limitations of an traditional IoT network by providing us with several key features like : Optimized for on-device machine learning, by addressing 5 key constraints: latency (there's no round-trip to a server), privacy (no personal data leaves the device), connectivity (internet connectivity is not required), size (reduced model and binary size), and power consumption (efficient inference and a lack of network connections).

What we learned

Keras Deep Learning Neural Network, Ngrok, Programming Alexa

What's next for Home Safety Using Fog Computing

To design Hardware/Software Components to incorporate our lite-weight system. Trigger safety events based on the prediction and classification like turning on sprinklers / Anti-inflammatory substances.