Graphite Parts,Gasket Graphite,Graphite Heating Elements,Graphite Exhaust Gasket Material CARBONS TECH&CREATION(HENAN) INDUSTRY CO.,LTD , https://www.moldedgraphite.com
Target learning based on deep learning
1. The necessity of target tracking technology
In recent years, intelligent video analysis technology has become a hot spot for security companies. Traditional security technologies are more concerned with the effectiveness of post-event verification. But with the popularity of high-definition cameras, 4K, HD, H.265, how to use these resources, how to make equipment "live", not just a machine that is activated after personal injury or property damage occurs, this is the more More and more security companies focus on the development. With video analytics, you can find out the anomalies in your video in a timely manner and respond in the first place, reducing losses. Among them, target tracking based on deep learning has become a hot topic.
2. Target tracking platform
Security companies are researching and developing the video analytics software platform. In order to make the data more accurate and free from interference from external factors, the algorithm based on deep learning is the key point. Take the Beijing and Pudong special software platform as an example and connect it to the military. The "integrated system platform" such as forest fire prevention, maritime surveillance, water conservancy and shipping can achieve the following functions.
(1) The 3D box selection highlights alarm information. Adopt sound+optical dual alarm mode. After receiving the front-end alarm, the preset sound alarm will be activated within 1 second. At the same time, the live image of thermal imaging and visible light, which generates the alarm, will be popped up, and will be added in the video image. The alarm position identification box highlights the alarm point information, and also displays the alarm position and distance information at the same time for user observation.
(2) Interactive electronic map and positioning display, based on three-dimensional GIS elevation data positioning. After receiving the alarm positioning data returned from the front end, the GPS position of the alarm point is rapidly and accurately located according to the GPS position of the front end, the position information of the pan/tilt, and the map 3D GIS elevation data, and the position of the alarm point is marked in the electronic map, and the positioning accuracy is determined. high.
(3) tracking and identification. The front-end device cooperates with the back-end software to automatically track the targets in the monitoring area, such as the trajectory of people, vehicles, and other activities.
(4) Radar linkage. The radar-camera warning linkage uses a "point by point" scanning technique to determine "point" targets through "point" scanning. The radar sends angle information back to the linked computer through the radar computer, and the linked computer compiles the radar “protocol†into the “protocol†of the pan-tilt and transmits it to the pan-tilt to control the rotation of the camera and accurately locate it, thus realizing the linkage between the radar and the camera.
The linkage between the radar and the camera can adjust the position of the camera, accurately locate the real-time position, view the scene in real time, further confirm the authenticity of the alarm, improve the accuracy of the alarm, and provide a real picture for the remote viewing site.
3. The development prospect of video analysis technology
With the advent of deep learning technology, the security industry's technology has taken a new step. Although current tracking based on deep learning targets goes beyond traditional security technologies, there is still a lot of room for optimization. For example, target tracking based on deep learning can not effectively deal with complete occlusion problems, and the human brain relies on its own strong prior knowledge to deal with these problems. The tracking algorithm can only learn the appearance of the target's current state, and does not have humans. The ability to calculate the brain still has a long way to go in deep learning.