The year 2023 is almost half way in and will enter it’s second half in a blink of an eye.
Thank you for your attention to SRIZFLY over the years.
“A journey of a thousand miles begins with a single step”.
Over the past two years, SRIZFLY has been continuously polishing self-developed flight simulation engine, maturing the flight simulator products day by day.
Today we announce development of content for UAV sensor simulation
Let's take a look at new features~
Sensor simulation
The SRIZFLY flight simulation engine can realize the environmental perception simulation of unmanned systems through optical sensor imaging simulation (including visible light, infrared and LIDAR), and provide RGB camera and its’ video stream output, LIDAR and its point cloud data real-time generation, which can be widely used in unmanned system algorithm verification.
· Visible light simulation
The camera simulation builds a three-dimensional model of the object based on the geometric and spatial information of the environmental object, and adds color and optical properties to the three-dimensional model through computer graphics based on the real material and texture of the object, including the simulation of monocular, binocular and fisheye cameras.
![300x200](http://v4-upload.goalsites.com/897/1686825395_图片1.png)
![300x200](http://v4-upload.goalsites.com/897/1686825413_图片2.png)
![300x200](http://v4-upload.goalsites.com/897/1686825432_图片3.png)
![300x200](http://v4-upload.goalsites.com/897/1686825485_图片4.png)
![300x200](http://v4-upload.goalsites.com/897/1686825505_图片5.png)
![300x200](http://v4-upload.goalsites.com/897/1686825527_图片6.png)
The SRIZFLY flight simulation engine generates high-precision infrared imaging maps and various infrared evaluation indicators in real time. The simulation engine has undergone multiple sets of test data verification and tuning, and has high precision and real-time calculation efficiency. The camera simulation uses the coordinate system conversion method to transform points in three-dimensional space into points on the image through the perspective relationship. Simulate the structure and optical characteristics of the camera lens, the data acquisition and processing of the camera sensor, the signal processing of the camera image, and the camera target-level recognition results of some integrated AI chips. The specific process is as follows:
![](http://v4-upload.goalsites.com/897/1686825647_图片7.png)
▲ Camera simulation process
![300x200](http://v4-upload.goalsites.com/897/1686825692_图片8.png)
![300x200](http://v4-upload.goalsites.com/897/1686825708_图片9.png)
LIDAR simulation simulates the working process of radar photoelectric transmission and reception. The laser beam intersects with all objects in the simulation scene, and the laser reflection intensity and noise of the intersection point are calculated based on the physical material type and properties of the intersection point.
The SRIZFLY simulation engine uses GPU-accelerated LIDAR simulation methods and RTX graphics real-time ray tracing technology to simulate LIDAR point clouds that are extremely close to the real world data.
The intensity of LIDAR reflection is affected by the distance of the obstacle, the angle of laser reflection, and the physical material of the obstacle itself. During the simulation, it is necessary to set up suitable physical materials for the scene resources, including various towers, flowers and trees, terrain, hydrology, obstacles, line corridors, etc. The SRIZFLY simulation engine extracts the reflection intensity model from the real radar scanning data to drive the simulation model, and obtains the reflection intensity and noise of the physical material under the current LIDAR through actual calibration.
![300x200](http://v4-upload.goalsites.com/897/1686825804_图片10.png)
![300x200](http://v4-upload.goalsites.com/897/1686825821_图片11.png)
Having scarce amount of available high-quality marking data in the power field, the use of virtual samples can be combined with a available manual marking samples to iteratively increase the learning method. In addition to numbers of actual training samples, possibility of modelling damages such as the self-detonation of insulators, or the absence of shock absorbing hammers, facilitates great expansion of the training set, resulting in improved image target recognition rate. This method plays a significant role especially for case with small amount of samples, but high demand for the recognition rate, and presents a good application prospects in different industries.
![300x200](http://v4-upload.goalsites.com/897/1686825909_图片12.png)
![300x200](http://v4-upload.goalsites.com/897/1686825925_图片13.png)
![300x200](http://v4-upload.goalsites.com/897/1686825941_图片14.png)
![300x200](http://v4-upload.goalsites.com/897/1686825967_图片15.png)
![300x200](http://v4-upload.goalsites.com/897/1686826169_图片16.png)
![300x200](http://v4-upload.goalsites.com/897/1686826191_图片17.png)