Fire Area Detection based on Convolutional Neural Network and Improved A∗ Path Planning

In this research, a new fire detection method and a new path planning algorithm are proposed. Traditional fire detection methods are based on fixed RGB models and they are not accurate enough under different circumstances. Traditional A∗ path planning algorithm always focuses on a minimum cost from a start point to an end point, so it is not suitable for complex environment. To solve fire detection problem, we use an object detection algorithm based on a convolutional neural network and trained it with real fire images to detect fire area in an image. For the path planning problem, an improved A∗ algorithm with new weight for different area and box blur method are used to ensure the output path is away from the obstacle. In the end of this paper, we simulate disaster environment in Unity3D and implement two algorithms to measure their performance.