Method and device for detecting a target object

Abstract

The invention relates to the technical field of pattern recognition, and provides a target object detection method and equipment. The method comprises the following steps that: according to an image feature vector and Agent historical action data, updating a current state; according to the current state, through the first DQN (Deep Q Network) of a Double DQN algorithm, obtaining a first expected value function value corresponding to each action; according to the first expected value function value and a decision parameter [Epsilon], adopting an [Epsilon]-greedy strategy to select a next action, and detecting the target object; and according to the execution result of the next action, adopting the second value function of a second DQN network in the Double DQN algorithm to evaluate the next action. By use of the target object detection method and equipment provided by the invention, the first expected value function value and the second value function of the Double DQN algorithm are independently adopted to select and evaluate the action, a possibility that the action is overhigh estimated can be effectively lowered, and detection performance is improved.

Description

technical field [0001] The present invention relates to the technical field of pattern recognition, and more specifically, to a target object detection method and device. Background technique [0002] At present, vision-based service robots are receiving more and more attention. The tasks of the robot service process include: target detection, navigation, and target capture. During the entire task process, target detection plays an important role. Once the detection target is not accurate enough, it will lead to the failure of the entire subsequent task. Therefore, the accuracy of object detection is crucial for service robots. [0003] In recent years, many object detection methods have emerged. In the past two years, some scholars have also used deep reinforcement learning for target detection. For example, Caicedo and Lazebnik used deep reinforcement learning to train Agents and deform the bounding box until it fits the target. Then Bueno et al. added a fixed hierarch...

AI Technical Summary

Problems solved by technology

[0004] However, the existing deep reinforcement learning methods for target detection are all based on DQN. DQN uses the same expected value function to select and evaluate an action, which can easily lead to overestimation of the selected action, thereby reducing The accuracy of target detection will also have a huge impact on the follow-up work of service robots

Images