This paper provides a computer vision based technique for detecting fire and identifying hazardous fire by processing the video data generated by an ordinary camera. While there are copious amount of publications in fire detection with images, distinguishing hazardous fire from non-hazardous is a problem that is still unsolved. The proposed technique uses the color and fluctuation characteristics of fire to detect it. Initially the algorithm locates regions of the video where there is motion; from these regions fire colored pixels are extracted and then wavelet transform is applied to confirm that the moving object is fire. The proposed technique tracks the rate of increment of fire region to distinguish between hazardous and controlled fire. Experimental results demonstrate that the proposed technique is effective in detecting the fire and classifying it as hazardous or controlled. The performance of the technique was observed in terms of true positive rate which was observed to be 85.57% while the average delay in detection of hazardous fire was 66.2 frames at a frame rate of 10 frames per second or 6.62 seconds.