Details

Autor(en) / Beteiligte

• Wu, Jun
• He, ZhongShi
• Yan, Kai
• Xie, HouLi
• Huang, QiCong
• Tang, Ying
• Tan, Haiyan

Titel

Key Points Detection Algorithm of Object Based on Full Convolution Network

Ist Teil von

• 2020 International Conference on Artificial Intelligence and Computer Engineering (ICAICE), 2020, p.158-162

Ort / Verlag

IEEE

Erscheinungsjahr

2020

Quelle

IEEE

Beschreibungen/Notizen

• This paper proposes a method for detecting key points of an object based on a fully convolutional network. The method for estimating all the key points of the object includes feature extraction without spatial information loss and regression of numerical coordinates. The first step mainly uses KPDA-Net to construct the AI model and output the feature map; the second step is to return the numerical coordinates according to the feature map; finally the algorithm design and experiment are completed. In this paper, we use the key point detection data set of All Tianchi competition to train and test the model. The key point detection accuracy and speed of the algorithm are (PCP mean: 81.3, t: 300ms). The advantage of this algorithm is that the output structure of the algorithm can be flexibly adjusted to be suitable for the key point detection of new object categories, so the algorithm has a strong migration ability.