Multi-local Feature Target Detection Method Based on Deep Neural Network

In application of video surveillance system, the algorithm of object detection is affected by occlusion easily, and the results on small object are not satisfying. This paper develops a multi-local feature object detection method based on deep neural network. The image is used as input to calculate the position and category probability of the object through a single network calculation, which improves the operating efficiency. The core of the method is to extract multiple local features of the target for detection. When the target is partially occluded, it can identify the target by the unoccluded patch. In addition, the high-level and low-level features in the convolutional network integrate to improve the detection effect on small targets. Experimental results show that the proposed method has a good effect on the detection of occluded targets and small objects.