Thrown object detection method and system based on semantic segmentation network

Abstract

The invention discloses a thrown object detection method and system based on a semantic segmentation network, and particularly relates to the field of road detection. The system mainly comprises an offline modeling module, an online background acquisition module and an online throwing judgment module. According to the method, the thrown object is indirectly judged through judgment of the moving target instead of direct judgment of the thrown object, the pre-detection rate of the thrown object is increased, and meanwhile, the preset background image is updated by adopting double backgrounds anda background updating strategy; and through similarity judgment of the candidate judgment area and the candidate judgment area and the preset background image, double judgment is carried out, so thatthe judgment accuracy is improved.

Description

technical field [0001] The invention relates to the field of road detection, in particular to a method and system for detecting spilled objects based on a semantic segmentation network. Background technique [0002] With the development of road traffic, the speed of inter-city transportation has been greatly increased. However, more and more freight vehicles have also caused many problems. In addition to frequent rear-end collisions and collisions, there are also vehicles to avoid falling cargo. Therefore, in addition to regulating driving behavior, the detection of accidental spills is also an important reference factor for traffic safety. [0003] Most of the existing detection methods for spilled objects on traffic roads use the traditional manual feature extraction method for detection, such as SIFT corner feature, LBP texture feature, etc. This method is affected by human factors. When changing the monitoring scene, the feature Insufficient generalization ability, weak...

AI Technical Summary

Problems solved by technology

[0004] In the actual traffic scene, the forms of spilled objects are various, including cardboard boxes, wood, iron blocks, plastic shells, rain cloths, tire skins, etc., but because the amount of such materials is too small, if the direct Undoubtedly, it is very difficult to carry out training and learning on objects. First, it is difficult to cover all types of spills. Second, there are few sample materials, so it is difficult to effectively generalize untrained objects. Therefore, the method of learning and training on objects is directly adopted. Difficult to implement effectively and unable to adapt to the current transportation environment

Images