An Adaptive Deep Neural Network for Detection, Recognition of Objects with Long Range Auto Surveillance

Automatic detection and recognition of predefined objects in a video stream of Long Range Surveillance Systems is multidimensional problem regarding the complexity of the background, moving sensor(s) and low probability pattern recognition of targets. An Adaptive Fuzzy based Network topology is proposed in this paper which is used in parallel with Deep Convolutional Neural Networks (DCNNs). In order to decrease the network training time and increase the trained model effectiveness, an Optimization while Training algorithm is also adapted to train and optimize the overall DCNN and parallel Fuzzy based Network. The tests showed that regardless of the noisy and variable background and the contrast quality, the proposed novel deep learning topology remarkably improves the detection&recognition rate while achieving fast and very qualitative decisions.