火灾严重危及着人类的生命财产安全，造成严重的后果。火焰作为火灾发生时一个显著特征，可以根据它的特性来对火灾的现象进行分析判断，进而对火灾进行预警，降低火灾的危害。视频监控系统的广泛应用和发展以及智能化技术的发展，使得利用视频信息来对火焰进行检测并报警显得实际起来。本文研究的方法与传统的火灾预防方法相比，它具有探测的范围广，识别率高的优点，同时对灾后的调查具有特别大的帮助。 本文基于AM3715设计了火焰传感器的硬件系统和软件系统，针对无人实时监管一些地方或者一些危险厂区，当火灾发生时，传感器系统能够有效的识别出火焰，给出报警信息，通知人类进行进一步处理。在本文论述的检测方法中，首先分析比较了帧查法和混合高斯背景方法，针对两种方法做了改进之后，选取改进的高斯算法来提取前景目标，接着对得到的结果进行滤波处理后，得到疑似火焰图像。本文通过对火焰图像进行统计分析，得出火焰的颜色特征，以此再作为进一步的判断依据，同时本文深入研究了火焰的角点的特征，将之作为判据，最后根据大量学者的一些研究，提出了基于亮度变化的累加器来描述火焰频闪的数学模型来进行最后的火焰决策。 本文构建了火焰检测的硬件平台，在devkit8500系统平台上实现了上述提到的各种方法，实现了视频采集、火焰识别、系统报警、视频传输等功能。在不同场景下进行测试，通过实例验证了系统的可行性和准确率，对视频检测火焰的产品化具有现实意义。

Fire is a serious disaster that threatens human production and bring a great threat to human life and property.The flame,as a notable feature when the fire broke out,it can be analyzed to determine the characteristics of the phenomenon of fire,and then do early-warning and reduce the fire hazard.The development of a wide range of technologies and intelligent video surveillance system, makes using the video information for flame detection and alarm seem real. Existing detection methods can not remove some of the complexity of the problem of interference environment, is likely to cause negatives and false results,the method of this paper compare with the traditional methods of fire prevention, the detection of a wide range of its high recognition rate, and it has a particularly big help when post-disaster survey. This paper designed hardware platform and software systems flame sensor for real-time monitoring based AM3715 in unmanned some places or some dangerous plant, when a fire occurs, the sensor system can effectively identify the flame, an alarm message notify Human process. This article discusses the detection method, first discussed the gray image binary method, and then after a thorough analysis of the background investigation method and Gaussian model modeling method for the frame difference method has been improved and found that it still can not meet system needs, the algorithm of the Gaussian model is more complex algorithm has been improved greatly simplifies the algorithm so that it can be applied to embedded systems.Next, the results obtained after the filtering process to obtain a suspected flame image. Based on the large number of flame image statistics, feature information obtained flames, as this further judgment. In this paper, in-depth study of the characteristics of the corner points of the flame, to use as a criterion to identify the video in flames. Finally, a large number of scholars, according to some studies, the establishment of real-time statistical analysis based on the strobe model is proposed based on the brightness change accumulator mathematical model to describe the flame flicker. And made corresponding experiments to validate our algorithm. This paper also built hardware platform of the sensor,it achieve a variety of methods mentioned above on the devkit8500 platform,and actualize the video capture, flame recognition、system alarms、video transmission and other functions. Finally test the sensor under different scenarios, validate feasibility and accuracy of the system, has practical significance for the products of video fire detection.