Feature Fusion Human Object Detection Algorithm

Due to the strong changes in the illumination intensity and the serious mutual occlusion of outdoor substation environment, it is difficult to detect human objects. In this paper, a Feature Fusion Faster SSD (Triple F-SSD) personnel object detection algorithm based on adaptive difference threshold for substation scene is proposed. The algorithm includes two parts: (1) Background Adaptive Difference (BAD) method based on the fixed background reference. This method utilizes the image difference method based on the fixed background reference to effectively remove the background information in the image. (2) Triple F-SSD personnel object detection algorithm. The algorithm adopts the feature fusion of the original image and the pre-processed image to retain the image context information while enhancing the features of the target area. According to different object scales, different feature maps are selected for processing, so as to improve the computational efficiency of the algorithm while enhancing the object features. Experimental results show that the object recognition accuracy of the proposed method can reach 92\% when it comes to person object detection in complex background of substation.