Weak Vehicle Target Recognition Based On The Mechanism Of Visual Cognitive

With the features of low cost and ease of use, the image-based vehicle identification technology has become the main technology of vehicle identification. Image information is complex and sensitive to environment conditions, which contributes to the poor robustness and weak adaptability of the image-based vehicle identification methods. Thus the vehicle identification, especially for weak vehicle targets, has become a challenging and significant job demanding to be solved urgently.Therefore, a robust vehicle identification model is established in this paper based on the principles of human visual perception. Moreover, the paper tries to build an associative mechanism model to obtain reliable recognition ability for weak vehicle target. Thesis detailed work is as follows:1. The proposal of the Driven-binary Vision Selective-Attention Model(DVSM) based on the human visual selective attention mechanism for vehicle recognition. The model improves the most influential visual selective attention model current--the Saliency model based on the bottom-up process involving in two aspects so as to realize the precise identification of the target vehicle. On the one hand, spectrum analysis method and significant degree function instead of multi-scale characteristic of the gaussian pyramid fusion improves the real-time performance of the model; On the other hand, the top-down upper task-driven process is introduced to guide the data-driven process of the Saliency model to realize the precise identification of vehicle target, which uses the most robust features of the vehicle target to establish two levels of the knowledge base. In the shape feature knowledge base, this paper proposed a segmentation mode based on the theory of gestalt perception to extract the shape feature set of vehicle objectives.2. The proposal of the Associative Mechanism Model for weak vehicle target identification. Firstly, the Green Neurons-interactive Associative Network(GNAN) is built consist of the associative mapping process and the inverse associative mapping process, which simulates the interaction function of the neurons in the neural signaling process using Nerve modulation function and Nerve interaction function to obtain faster convergence speed. On the basis of the GNAN, this paper establishes a hierarchical Associative Mechanism model consist of the association produce phase, the association matching phase and the comprehensive analysis phase based on the mechanism of the WHAT-Path in the visual cortex, which obtains excellent recognition ability for weak vehicle target.Through simulation experiments, the recognition rate of DVSM reaches90.4%and the false recognition rate is4.9%for clear vehicle targets, while it is76.4%and4.8%for comprehensive objects set consist of both clear vehicle targets and weak vehicle targets; After the introduction of the Associative Mechanism Model, the comprehensive recognition rate becomes88.5%while the comprehensive error recognition rate is4.9%. Experimental results show that the DVSM owns excellent identification ability and robustness for clear vehicle target and the Associative Mechanism Model will improves the recognition rate significantly for weak vehicle target and will not change the false recognition performance of the system.