China News Service, October 28. According to a report from Hong Kong Sing Tao.com, there has been a surge in hiking accidents recently. In order to improve search and rescue efficiency, the Hong Kong Fire Services Department has cooperated with a university team to develop drone photo analysis software to help identify similar humanoid objects in photos. Save the time of looking at photos with the naked eye in the past, and speed up the search for missing persons for rescue.
Assistant District Chief (Special Service Support Team) of the Hong Kong Fire Services Department, Wang Dehui, said that mountain climbing rescues can be divided into two categories. The first is that the seeker can clearly state the location, and the second is the accident of uncertain location, usually hiking alone, or when relatives and friends find it. The victim was missing and reported to the police. There are more than 10 requests for these requests every year. After receiving the information, the fire department will send personnel to go up the mountain and assign a large number of search routes. .
Wang Dehui pointed out that the Fire Services Department has cooperated with a local university innovation and technology group to successfully develop the “Artificial Intelligence Mountain Climbing Rescue Drone Photo Analysis Software” (referred to as the “software”), which has been piloted.
It is reported that the drone can take about 1,500 photos in the range of 1 km by 1 km. It takes 8 hours to inspect each photo with the naked eye, and the “software” can be completed in 2 hours. If the built-in functions of the computer are improved, the analysis time will be faster. .
Wang Zhengqi, Senior Land Surveyor (Information Technology Management Team) of the Fire Services Department, said that the “software” uses a “deep learning model”, the program will record the data of the photos taken in the past, automatically divide the photos into multiple small cells, and then according to the images in the cells The color, shape, environment contrast and size are used to determine whether there are similar humanoid objects, and then the small grids are recombined to calculate the probability of similar humanoid objects in the photo.
Wang Zhengqi continued to say that the system is equipped with “one-button imitation flight”, which can pre-set the drone route, press one button to automatically fly and take pictures, and can also use 5G streaming technology and drone footage for real-time Live broadcast, understand the situation in real time, and then make action deployment. He emphasized that the system will continue to be optimized, and is currently working on identifying the characteristics of the missing, such as clothing, rucksacks, etc., as well as supporting thermal imaging photo analysis.