Occluded Vehicle Target Detection Method Based On Conditional Generative Adversarial Siamese Network

In vehicle target detection process, due to the existence of occlusion, the feature of the vehicle target to be detected will be missing. So the detection accuracy will be reduced. Therefore, we propose a conditional generative adversarial siamese network for vehicle detection. The occluded vehicle network is divided into two parts: occluded feature generator and discriminator. Firstly, the random occlusion is generated for the data set as the input of the model. Then, the pooling feature of the occluded vehicle is restored by the generator, and the pooling feature of the restored occluded image is distinguished from the pooling feature of the unoccluded image by the discriminator. The siamese network is used to extract features from reconstructed images, which further improves the representation ability of the model. Finally, the experiment results show that the proposed model can accurately detect the occluded vehicles compared with other state-of-the-art methods.