Applications of Artificial Intelligence in Fire & Safety | by Vedant Kumar | Dec, 2020
[ad_1]
Being a student secretary of the Fire and Security Association of India at my college, I have conducted various surveys and workshops on methodologies to avert the dangers caused due to fire. As an AI enthusiast, I have always aspired to come with innovative solutions to tackle such issues. My experience in the fire and safety domain with competence in building AI solutions motivated me to think of possible solutions to tackle problems faced by the fire department.
And so, in this article, we’ll look at how techniques like computer vision, machine learning, and deep learning can help the front line workers to make more informed decision in a stressful situation. The data coming from the sensors would have a positive impact — both in terms of rescue time and the life of the victim as well as the fire respondent.
A common problem faced by the fire officials when a rescue operation is underway in a place that has engulfed with flames and smoke is poor visibility. It becomes difficult to notice objects like a door, staircase, or any obstruction in their path of rescue, causing a delay in executing the rescue operation. Such issues are concerning, especially when a life is at stake.
To solve this issue, a simple computer vision technology is employed. By mounting a camera on top of the helmet, data from the camera is used to enhance the vision of the fire officials. The flow of the overall process would be as follows. The image frames from the camera are sent to a processor for computing simple matrix operations. The processor will carry out edge detection and contour detection operation using various filters like a Sobel operator. The output from the processor will be published in AR glasses of the fire official. The view of the AR glasses would be similar to the figure above. With this simple algorithm, the fire officials can discern the surrounding efficiently. This makes it easy for the firefighters to identify objects around them, thereby expediting the entire process of salvation.
Apart from vision-related problems, the heat from the flames can cause suffocation, seizures, and even heart attacks. So much so that most cases of fire official deaths are due to one of the above-mentioned threats. To mitigate such dangers, an IoT (Internet of things) based solution can be put in place. Sensors like oxygen level detectors are used to check if the oxygen level of the firefighter is in the normal range. ECG sensors can understand the extent of physiological arousal that someone is undergoing to better infer about someone’s psychological state. Body temperature sensors monitor the change in temperature of the human body to ensure that the temperature range is within the safe bracket. With a plethora of data, a recommendation system warns the official in case of an anomaly in the data. A machine learning model can be used to classify the data from various sensors in the range-safe, danger, and needs immediate attention. Many algorithms can be used for this purpose, like decision trees, random forest, ANNs, K-means clustering, and so on. The flow of data is as follows. The data from sensors is given to the processor. The machine learning model gives a probabilistic output or a categorical output. The output data can be used by the supporting team to warn the officials in case of danger by communicating with them over a call.
AUDREY strands Assistant for Understanding Data through Reasoning, Extraction, and synthesis. It is an AI solution developed by NASA. In the previous section, we have discussed how some sensors can alleviate a stressful situation by understanding the data from the body of the respondent. However, AUDREY analyzes surrounding data and recommends the safest plan of action. For example, if AUDREY senses a high level of toxic gases like carbon monoxide, it can caution the respondent suggesting him to be more vigilant. Using artificial intelligence, AUDREY can collect data on temperatures, gases, and other danger signals and guide a team of first responders safely through the flames.
Two approaches are used to fight forest fires using AI- image-based and sensor-based.
Image-based approach:
In an image-based approach, a convolutional neural network is trained to detect the fires. The process involves preparing the dataset, annotating the dataset, training the model, and testing the model for validation. The deep learning model is usually deployed on drones which are used for surveillance purposes to detect the presence of wildfires. The task is very similar to that of an object detection model. Once the model has learned the features of the flames from the data and its annotations, it can be used for the detection purpose. The most commonly used model is MobileNet because it offers good accuracy with less computational complexity owing to the use of depth-wise convolution. This can be further extended to detect the severity of the incident by using the k-nearest neighbor algorithm. This approach is similar to what the famous YOLO model uses. The drones detect these flames and warn the authority to take the necessary action. In this way, deep learning is used to mitigate the extremity of the catastrophe.
Sensor-based approach:
In a sensor-based approach, a variety of sensors present in the forest make a cumulative prediction about the occurrence of forest fires. Sensors are used to measure the amount of carbon dioxide, hydrogen sulfide, carbon monoxide, and oxygen in the atmosphere. This data coupled with variations in the surrounding temperature along with the change in humidity level is used for the detection of wildfires. In this approach, the machine learning model is trained on millions of datasets and is competent to make accurate predictions. When an anomaly is detected by the ML model, it raises an alarm to warn the forest officials. Again, classification models like the random forest can be used to detect forest fires. An advanced application of the same concept would be to predict the occurrence of the next forest fires. This is a time series problem and can be solved using deep learning models like Long Short-Term Memory (LSTM) or recurrent neural network (RNN).
Therefore, AI has an enormous potential to grow in the domain of fire and safety. These are just a few applications that can be employed to protect the environment and the lives of the people from a mishap.
Read More …
[ad_2]