Research On Fire Recognizing Algorithm And Its Implementation On DSP

Prevention and reduction fire disaster is a worldwide problem. Although various fire preventions are adopted by countries, serious personal casualty and property losses are caused by fire. Especially, the problems of great space and field fire prevention are increasing outstanding with expanding human activities. Great efforts and resources are needed by artificial fire prevention of these spaces. Due to the limits of performance of traditional fire detections, which have realized automotive fire prevention, they are not suitable for large range and severe environment. Fire detecting system is able to overcome the limit detecting range and poor environmental adaptability of traditional fire detecting system.In this paper, fire detecting system based on DSP is designed and implemented with ICETEK-DM6437-B-KIT of Realtime Technology Co, Ltd, Beijing as hardware platform. The system is composed by pre-processing module of fire image, splitting module of suspected fire area, fire recognition and fire tracking module. The main works of this paper are as follows:(1) Simulation and selection of algorithms. Common algorithms of image pre-processing, area splitting and target tracking algorithms are simulated and analyzed in this paper. Suitable algorithms are selected by analysis of efficiency and complexity.(2) Improvement and optimization of algorithms. In this paper, background difference and Kalman prediction are improved. Owing to the labile working environment of fire detecting system based on video, stable background image is hard to obtain. Establishment method of background based on mean value of each frame, which ignores too big and small pixel value and obtain quadratic average value, is proposed on the basis of traditional background difference. Moreover, static search window of traditional Kalman prediction is not appropriate to fire tracking according to serious and irregular changing fire. Variable research window of Kalman prediction based on tendency is proposed.(3) Transplant and optimization of algorithm code. On full consideration of hardware and software features of TMS320DM6437, algorithm code is replanted and optimized, including memory, code and compiler optimization.(4) Performance test. After realized the function of system, the performance of the system is tested, including efficiency and effectiveness test.Performance test shows that the efficiency and effectiveness fulfills requirement and provide theoretical and technical foundation.