A Novel Smoke Detection Method Using Support Vector Machine

1. A Novel Smoke Detection Method Using Support Vector Machine By Mounica

2. Purpose of the paper • Introduction • Technical Aspects • Results • Conclusion • Drawbacks and Future Work Content:

3. Early Fire detection is a big problem in large facilities like port facilities, large factories and power plants to keep them safe, due to its large harmful effect to surrounding areas. • In such areas Smoke is an important and useful sign to detect fire. • The Paper aims on a novel and robust smoke detection methods based on image information. Purpose Of The Paper

4. Previous approaches of the smoke detection, using pixel and block-based, or color-based image processing methods detect edge or contour information of smoke. • In this paper novel smoke detection method is being presented with the image information obtained from the surveillance cameras. • In the pre-processing stage moving objects are detected as smoke regions. • Texture analysis and non-linear classification with SVM to extract smoke regions in image sequences is used here. Introduction

5. Pre-Processing: • This is an important step before we proceed towards the smoke detection method. • In this step moving objects are detected from the images as smoke regions with the help of some methods • Subtraction and Accumulation : • A subtracted image frame is written as 𝑔(𝑡)=𝑓(𝑡)−𝑓(𝑡−1) • Smoke is not clear in the subtracted image • So, we use ℎ(𝑡) which combines two subtracted images, i.e. ℎ(𝑡)= ∣𝑔(𝑡)+𝑔(𝑡−1)∣ Technical Aspects

6. Technical Aspects (contd.…..)

7. Binarization and Morphological operation: • Binarization and Morphological operations are done to remove noise like regions in the binary images. • Binarization is the process of converting an image into black and white image by choosing a threshold value. That is ℎ(𝑡) is converted to 𝑏(𝑡) binary images. • Otsu’s method is used to choose a threshold • Morphological operations probe an image with a small shape or template called a structuring element. The structuring element is positioned at all possible locations in the image and it is compared with the corresponding neighbourhood of pixels. Some operations test whether the element "fits" within the neighbourhood, while others test whether it "hits" or intersects the neighbourhood.A morphological operation on a binary image creates a new binary image in which the pixel has a non-zero value only if the test is successful at that location in the input image. Technical Aspects (contd.…..)

8. Extraction of Feret’s regions: • In this to determine the shape and position of the moving objects Feret’s diameter is extracted as a circumscribed rectangle whose horizontal and vertical lengths are Feret’s diameter. • We obtain the position and the approximated shape of the object as the rectangle thgis is called Feret’s region F(𝑡;𝑖). Technical Aspects (contd.…..)

9. Texture Features: • After detecting Feret’s region texture patterns of smoke are collected as feature vectors. • The texture feature is stored in a co–occurrence matrix, so it does not depend on the visible size of smoke in images. • So, texture features are very useful for the method of detection. Technical Aspects (contd.…..)

10. Methodology: • SVM(Support Vector Machine): • A Support Vector Machine (SVM) is a discriminative classifier formally defined by a separating hyperplane. In other words, given labeled training data (supervised learning), the algorithm outputs an optimal hyperplane which categorizes new examples. In two dimensional space this hyperplane is a line dividing a plane in two parts where in each class lay in either side. • We use texture features as input vector to SVM • We use non–linear kernel (RBF kernel) of SVM as 𝐾(𝑥𝑖,𝑥) = exp(−𝛾∣∣𝑥𝑖 −𝑥∣∣2) , where 𝑥𝑖 ∈ ℝ14 is a texture feature vector of a Feret’s region. Technical Aspects (contd.…..)

11. To train the SVM, we prepare manually selected Feret’s’ region of smoke, which we call ideal smoke. • Using trained SVM, we discriminate whether Feret’s’ regions are smoke or not. If the SVM’s output is smoke, we set the label 𝑙𝑔(𝑥,𝑦;𝑡) of the point (𝑥,𝑦) in the Feret’s’ region to be 1 and else to be 0. 𝑙𝑔(𝑥,𝑦;𝑡)= and 0(𝑒𝑙𝑠𝑒) • In real-time situation along with smoke there may be many moving objects as noise in the image. • To obtain the accurate result of the smoke detection, we consider to accumulate the labeling results 𝑙𝑔(𝑥,𝑦;𝑡) with SVM about time. The accumulation is defined as follows: 𝒜𝑔(𝑥,𝑦;𝑡)=𝑙𝑔(𝑥,𝑦;𝑡) • When there exist a point (𝑥,𝑦) whose 𝒜𝑔(𝑥,𝑦;𝑡) is higher than the manually selected threshold, all Feret’s regions which contain that point discriminated as smoke at time 𝑡 is extracted. Technical Aspects (contd.…..)

12. Experiment 1: Results

13. Experiment 1(contd…): Results

14. Results

15. Results:

16. Thus the paper presents the novel smoke detection method based on the texture analysis and the support vector classifier. Focusing on the image information of smoke as the texture pattern, which does not affect the size of smoke in the images. • For the evaluation the method, we examine it with some examples of image sequences. In the experiments, there exist smoke and other moving objects which are considered as obstructions in image sequences. Experimental results show the effectiveness of our method under general conditions. Conclusion

17. The method is not evaluated for different types of smoke. • To use this method in real situation, we must compile the smoke detection system with other systems. • One way is to combine our method with multi camera system to cover the wide target area. Drawbacks and Future work

18. Thank You!!!