Spatio-temporal information and deep network-based monitoring video object detection method

Abstract

The invention discloses a spatio-temporal information and deep network-based monitoring video object detection method. The method comprises the steps of collecting and labeling a data set and traininga deep convolution neural network; extracting robust multi-scale depth characteristics by utilizing the trained deep convolution neural network; extracting a moving target candidate region based on multi-scale depth characteristics; predicting the position of a target in a next frame according to detection results in the previous and latter frames of a video; carrying out RoI normalization on a motion target candidate region and a predicted candidate region, subjecting characteristic vectors to classification and regression, and obtaining a preliminary detection result; based on the motion and prediction information, finely adjusting the obtained preliminary result and further accurately detecting the result. According to the invention, the rich spatio-temporal information contained in the video is comprehensively considered, so that redundant candidate frames are greatly reduced by means of motion and prediction. Meanwhile, the problem that a single frame-based detection result is not stable is solved. Compared with other region-based detection methods for target detection, the time accuracy and the detection accuracy are both improved.

Description

technical field [0001] The invention belongs to the technical field of computer digital image processing and pattern recognition, and in particular relates to a monitoring video object detection method based on spatio-temporal information and a deep network. Background technique [0002] Today, a large number of cameras in cities can be used to capture uninterrupted visual surveillance information of important areas. It plays an important role in urban informatization, urban intelligent transportation and urban security. According to IMS Research statistics, in 2016, the shipment of urban surveillance cameras reached 138 million pieces, and the amount of surveillance videos generated every day reached thousands of petabytes. At present, the surveillance video business has entered the era of data gushing. For example, Shenzhen currently has more than 1.34 million cameras. Currently, the use of cameras in Shenzhen has accounted for 50% of the total number of criminal cases so...

AI Technical Summary

Problems solved by technology

In addition, due to the influence of light, distance, occlusion, complex background and other factors, traditional methods can no longer meet the growing demand in terms of detection time and performance

Images