基于轻量化SSD算法的行人目标检测

Pedestrian Target Detection Based on Lightweight SSD Algorithm

行人目标检测作为一种重要的目标检测应用,在自动驾驶,视频监控等领域取得了广泛的关注。但是由于基于深度网络的算法模型规模较大,需要强大的GPU算力支撑,导致该检测算法在嵌入式平台应用空间受限问题。现使用MobileNetV2轻量化网络代替传统的VGG网络进行特征提取,对SSD(Single Shot Multi-Box Detector)模型进行优化,实现减小模型规模,加快检测速度的目的。同时,通过减少筛选区域候选框以及选用自适应极大值抑制方法排除重叠检测框,提高了检测的精度和速度。在Caltech和VOC2007数据集上与其它流行的检测模型进行了对比实验。结果表明,上述模型检测准确率为71.8%,远优于其它模型大小相似的检测算法,为SSD模型在嵌入式平台的应用提供理论支撑和参考价值。

Pedestrian target detection, as an important target detection application, has gained widespread attention in areas such as autonomous driving and video surveillance. However, due to the large scale of the deep network algorithm model, which needs the support of powerful GPU computing power, the application space of this detection algorithm in embedded platforms is limited. In this paper, the MobileNetV2 lightweight network is used to replace the traditional VGG network for feature extraction, and optimize the SSD(Single Shot Multi-Box Detector) model to achieve the purpose of reducing the scale of the model and speeding up the detection speed. At the same time, the detection accuracy and speed are improved by reducing the candidate frame of the screening area and selecting the adaptive maximum suppression method to eliminate the overlapping detection frame. Finally, the Caltech and VOC2007 data sets are compared with other popular detection models. The results show that the detection accuracy of the above detection model is 71.8%,which is far better than other detection algorithms of similar size, and provides theoretical support and reference value for the application of the SSD model on the embedded platform.