面向嵌入式设备的高实时微小目标跟踪检测方法

传统主流目标检测算法在嵌入式平台无法兼具高实时性与高准确性，难以应用于边缘智能等领域。为解决微小目标跟踪检测在嵌入式平台实时应用的瓶颈，提出一种高实时微小目标跟踪检测方法。利用轻量化神经网络的骨干网络和路径聚合网络，对整体网络进行针对化的剪枝优化，同时深度融合相关滤波算法，提升针对微小目标跟踪检测的准确度和速度。在3D物体场景渲染器自建的军事微小目标数据集上的实验结果表明，在100像素的极小目标跟踪识别中，与DarkNet53-CSP方法相比，该方法检测精度大幅提高，在400~10 000像素的微小目标识别跟踪中，检测精度与检测速度优于DarkNet53和ResNeXt50+CSP等算法。

Highly Real-Time Method of Small Target Tracking and Detection for Embedded Device

Traditional mainstream target detection algorithms cannot provide both high real-time performance and high accuracy on embedded platforms, and thus have limited application in the edge intelligence field.In order to solve this real-time application bottleneck on embedded platforms, a highly real-time small target tracking and detection method is proposed.This method employs the backbone network of a lightweight neural network and a path aggregation network to prune the overall network for optimization.At the same time, it deeply integrates correlation filtering algorithms to improve the accuracy and speed of small target tracking and detection.The method is tested on a self-made small military target dataset built by using the 3D object and scene renderer.The experimental results show that in the case of 100 pixel small target tracking and recognition, the proposed method greatly improves the detection accuracy compared with Darknet53-CSP.In the case of 400~10 000 pixel small target tracking and recognition, the proposed method exhibits higher detection accuracy and speed than DarkNet53 and ResNeXt50+CSP.