Abstract:

Fast and accurate pedestrian and vehicle detection in UAV images is a meaningful but challenging task, which is widely used in military reconnaissance, traffic control and remote area rescue missions. However, UAVs are small mobile devices with limited memory and computing power, it is a real?challenge?for?real-time object detection on a UAV platform. Aiming at the problem that the SSD algorithm is too large and the operation memory consumption is too high to run on the UAV platform, a lightweight backbone network is carefully designed which reduces the number of channels and convolution of the original network, so as to reduce the number of parameters of the network. In addition, aiming at the problem of small object and complex scene in UAV scene, a lightweight receptive field block is proposed to enhance the feature representation ability, and the detection accuracy?of small object is further improved by combining context information module. Experimental results show that the proposed algorithm has high accuracy and real-time performance in the tasks of pedestrian and vehicle detection from UAV images.

Key words: Unmanned Aerial Vehicle（UAV） images, Single-Shot Detector（SSD）, object detection

摘要：

在无人机图像中快速准确地检测行人和车辆是一项有意义但又极具挑战的任务，其广泛应用于军事侦察、交通管制以及偏远地区救援等任务中。然而，由于无人机属于小型移动设备，其内存和计算能力非常有限，使得如何保证其检测实时性一直是难题。针对SSD算法模型过大、运行内存占用量过高、很难在无人机设备上运行的问题，精心设计了轻量级的基准网络，通过削减原始网络的通道数目以及卷积数目来降低网络的参数量；针对无人机场景下目标小、场景复杂等问题，提出轻量级感受野模块来增强网络特征表示能力，并结合上下文信息来进一步提高小型目标的检测精度。实验结果表明，提出的方法在基于无人机的行人与车辆目标检测任务上有较高的准确性和实时性。

关键词: 无人机图像, SSD, 目标检测