Abstract: People counting based on surveillance camera is fundamentalfor analyzing behavior of counting,resource optimization and resource allocation,modern security and protection,collecting commerce information as well as intelligent management.Therefore,it has significant meaning of study and application value.Recently,technology of digital image processing and theory of deep learning are constantly improved and developed,extremely promoting the study of people counting based on surveillance camera.However,there exist some problems,such as low accuracy of people counting and time-consuming of high definition,which are unable to be solved.In the wide range of object scale,accuracy of people counting method based on object detection decreases significantly.Aiming at this problem,this paper proposed a people counting method based onadaptive overlapping segmentation and deep neural network.The idea of this method comes from attention mechanism,and makes full use of information of the scales and numbers of head object in overlapping segmentation.The experimental results show that the adaptive overlapping segmentation algorithm can combine existing object detection model based on neural network.What’s more,compared with the method of counting people by directly using object detection model based on neural network,the combination algorithm of adaptive overlapping segmentation and deep neural network can greatly improve the accuracy of people counting.

Key words: Adaptive overlapping segmentation, Deep neural network, NMS, Object detection, People counting