Moving object detection for surveillance video frames using two stage multi-scale residual convolution neural networks

The human visual system's strongest suit is its ability to detect and follow moving objects. The rise of video-based applications like surveillance, traffic monitoring, military security, robot navigation, etc. may be attributed to the widespread availability of high-quality cameras. Strong object tracking remains a formidable obstacle, despite the many methods that have been developed so far. Thus, there are a number of issues that need to be addressed, including moving backdrops, foreground objects during training, variations in lighting, and occlusion. This paper explores visual tracking of an unknown entity that undergoes drastic visual transformations and often enters and exits the field of view of the camera. Detecting moving objects from the surveillance video data is a tedious task using the traditional techniques. In this paper a two-stage moving object detection methodology is proposed using Multi Scale Residual Block based Convolution Neural Networks (MSRB-CNN). The first stage is designed with a sequence of convolution layers with the CNN layers to extract the feature maps of interested regions of frames. In the second stage the refinement of feature maps to segment the foreground objects of the frames using the couple of MSRB layers. To increase the resolution of the output feature maps de-convolution layers are used.