377 JURNAL DARMA AGUNG, Vol. 31, No. 1, (2023) April : 377 - 382 SMOKE DETECTION ON CNN BASED VIDEO SURVEILLANCE SYSTEM By: Faris Rai Fadhil 1) Ari Purno Wahyu Wibowo 2) Universitas Widyatama, Bandung 1,2) E-mail: faris.rai@widyatama.ac.id 1) ari.purno@widyatama.ac.id 2) ABSTRACT Forest fires are a serious problem that can cause extensive forest land and plantation areas to be damaged, this damage not only disrupts the habitat but the ecosystems in the forest, several studies have made an experiment to prevent forest fires, one of which is by using the help of electronic sensors installed in forest areas, this sensor works chemically by detecting heat or a change in the composition of the atmosphere present in the air and room temperature, from these changes the data is sent to the central station and a fire will be predicted, this method has a weakness including the number of sensors installed is very limited and does not allow it to be installed in a large area and the sensor may be damaged and lost as well as the use of a limited power source, causing readings to become disruptive and less than optimal. The solution to this problem is to use a UAV or drone technology which is felt to be very effective and quickly moves areas to scan a very large area, while the fire detection process uses an image processing method based on a computer vision algorithm, this method reads forest fire data and calculates the area directly. and can be used to detect the impact of damage caused, the accuracy of this image processing system depends on whether it is clear or not the reading of the data captured by the UAV predicts an accuracy of up to 80%. Keyword: Forest Fires, Image Processing, Computer Vision 1. INRODUCTION In general, the design of a fire detection system based on video-based images with GPU (Graphic Processing Unit) support, this system works by reading video frame data, each video frame will be read by analyzing the model of the stages of reading the model, namely for Adaptive- GMM, making color tables, processing video frames, motion detection with Adaptive-GMM, fire color segmentation and finally a process for detecting fire from a combination of the results of motion detection and fire color segmentation and displaying the detection results on the screen. GPUs are used in methods that require high computations such as color conversion to grayscale color space and HSV of color changes taken from motion detection and fire color segmentation.