An Improved SSD Algorithm and Its Mobile Terminal Implementation
Enzeng Dong, Yao Lu, Shengzhi Du · 2019
Aiming at the problem of low real-time detection and poor portability of SSD (Single Shot MultiBox detector) algorithm in moving end, a real-time target detection algorithm based on improved SSD is proposed. The improved algorithm uses MobileNet which merges BN (Batch Normalization) layer as the basic feature extract network to compress the network model size and improve speed of target detection. In the data preprocessing stage, a set of data augmentation method is designed to ensure the accuracy of target detection. For practical application, the algorithm transplantation is carried out on Android mobile phone, and the target detection system of mobile terminal is realized. After experiments, compared with SSD, the proposed algorithm improves the detection speed and reduces the memory footprint on the basis of ensuring the detection accuracy, and can basically achieve real-time target detection at the mobile end.