Research On Impact Point Localization Algorithm Based On Salient Target Detection And Binocular Vision

Today,with the ever-increasing development of military weapon force reform and innovation,as the most important environment of military training,the need to use automatic reporting system of high-precision target to improve the efficiency of military training is becoming increasingly important.Existing algorithms for locating impact points using sound signals and Global Positioning System(GPS)have problems such as poor robustness,low positioning efficiency and redundant resources.Using the image signal to locate the point of impact,the method has low environmental requirements,easy equipment installation and high algorithm robustness.With the development of artificial intelligence,the application of deep learning technology has become more and more extensive.This article uses the deep learning-based salient target detection algorithm,combined with the binocular vision positioning system,to study the positioning coordinates of the shooting range impact point.In this paper,a binocular vision localization algorithm based on salient target detection is established based on the relationship between the pixel coordinates of the image impact point and the actual impact point.First,based on the pixel coordinates of the bullet impact point of the image,this article uses the salient target detection network to extract the salient area of the projectile flame.According to the requirements of the algorithm for detection accuracy and detection speed of the projectile flame,the original salient target detection network is introduced.The mechanism,pyramidal pooling modulus,and depth-separable convolution are used to improve the model to improve the saliency map detection accuracy and model inference efficiency;second,the salience map pre-processing The module is used to extract the salience map coordinates from the salience map,and select the appropriate coordinates The camera performs the subsequent calculation of the impact point coordinates;finally,using the binocular vision system based on the imaging angle,according to the pixel coordinates of the image impact point combined with the angle of the optical axis of the image of the point of auxiliary calibration of the shooting field,the binocular camera impact point imaging relationship is obtained simultaneously.The coordinates of the shooting field's point of impact.Based on the pixel coordinate part of the image impact point,the error of the improved algorithm proposed in this article in the precision of the pixel coordinate of the image impact point is reduced from the original coordinate error of 10.77 pixels to5.92 pixel coordinate errors;the size of the model is reduced from the original 348.5MB to 275.3MB,the algorithm Inference speed is increased by 1.7 ips,which improves the efficiency of the algorithm's inference.For the binocular vision positioning system based on the imaging angle in this article,compared with the traditional binocular vision positioning method,the bullet impact point positioning error is reduced to about 5 m,which improves the positioning accuracy of the bullet impact point of the shooting range,and can meet the project of positioning the bullet impact point of the shooting range.