Object detection and location generally require multiple sensors and multiple deep networks to achieve. For a specific application scenario, this paper proposes an algorithm based on semantic segmentation to display object detection and location at the same time. Due to the small size of the target in the image, especially the vertex of the target occupies very few pixels, the HRnet semantic segmentation algorithm with higher resolution is adopted. Due to the unbalanced class distribution and the small number of training samples, data augmentation and Lovasz Softmax loss function are used in a targeted manner. After experimental verification, in this data set, HRnet is superior to other algorithms. And by using data augmentation and Lovasz Softmax loss function, the accuracy and IoU have been greatly improved, especially the target vertex area with few samples.
Research on Object Detection and Location Using Semantic Segmentation
Lect. Notes Electrical Eng.
2021-12-14
10 pages
Article/Chapter (Book)
Electronic Resource
English
Privacy preserving anomaly detection using semantic segmentation
European Patent Office | 2021
Object Detection and Semantic Segmentation for a Small Low-cost Vehicle
Springer Verlag | 2021
|