Pedestrian re-recognition method based on retinex algorithm and convolutional neural network

Abstract

The invention discloses a pedestrian re-recognition method based on a retinex algorithm and a convolutional neural network. According to the method, a video frame sequence in a video database is extracted; the convolutional neural network is constructed, and a pedestrian network model is obtained through training; the trained network model is used to find out pedestrians from the video frame sequence; the retinex algorithm is used to perform image enhancement on the pedestrians; the enhanced pedestrians are inputted into the convolutional neural network, and the depth characteristics of the pedestrians at different levels are extracted; and classification is performed through the softmax classifier of the last layer of the convolutional neural network, so that a final matching similarity is obtained. Problems such as illumination change and shadow coverage in a real scene are fully considered; before recognition is performed, the retinex enhancement algorithm is introduced to simulate a human visual system, so that an image can be closer to what human eyes see, and therefore, a recognition effect can be effectively improved; and an end-to-end pedestrian re-recognition method is adopted, pedestrian detection and pedestrian recognition are combined through using the same convolutional neural network, and therefore, the alignment problem of pedestrian labels can be solved.

Description

technical field [0001] The invention belongs to the technical field of pattern recognition, and in particular relates to a pedestrian re-identification method based on a retinex algorithm and a convolutional neural network. Background technique [0002] Pedestrian re-identification is a hot research topic in the field of computer vision in recent years. Traditional pedestrian re-identification separates pedestrian detection and pedestrian recognition. First, the DPM or ACF algorithm is used to detect pedestrians, and the detected pedestrian bounding boxes are cropped for later use. Pedestrian recognition, this method is difficult to use in real scenes, and it is difficult to achieve complete alignment between the pedestrian detection area and the pedestrian recognition area, resulting in low pedestrian recognition. [0003] A popular method for pedestrian recognition is to use convolutional neural network to extract the picture features of pedestrian pairs for feature matchi...

AI Technical Summary

Problems solved by technology

The traditional neural network has a strong nonlinear fitting ability, can map any complex nonlinear relationship, has strong robustness and memory ability, and uses the low-level features extracted by the neural network, such as color texture empty structure features, etc., These features are easy to calculate and relatively reliable, but these underlying features do not have good semantic expression ability and are not closely related to the appearance of pedestrians

When there are changes in illumination and pedestrian rotation angles, these features often cannot express the characteristics of pedestrians well, resulting in deviations in recognition

Images