CN111652181B - Target Tracking Method and Device And Electronic Equipment - Google Patents > 자유게시판

Legal status (The legal standing is an assumption and is not a legal conclusion. Current Assignee (The listed assignees may be inaccurate. Priority date (The priority date is an assumption and isn't a legal conclusion. The appliance discloses a goal tracking method, a target tracking device and electronic equipment, and pertains to the technical discipline of artificial intelligence. The tactic comprises the following steps: a primary sub-community in the joint monitoring detection community, a first feature map extracted from the goal feature map, and a second characteristic map extracted from the target characteristic map by a second sub-community within the joint monitoring detection community; fusing the second characteristic map extracted by the second sub-network to the primary function map to obtain a fused function map corresponding to the first sub-network; buying first prediction information output by a first sub-community based mostly on a fusion feature map, and acquiring second prediction data output by a second sub-community; and figuring out the current position and the movement trail of the shifting goal within the goal video primarily based on the primary prediction information and the second prediction info.

The relevance among all the sub-networks that are parallel to one another might be enhanced via feature fusion, and the accuracy of the decided position and motion path of the operation goal is improved. The present software relates to the sphere of artificial intelligence, and in particular, to a target tracking method, apparatus, and electronic machine. In recent times, synthetic intelligence (Artificial Intelligence, AI) know-how has been widely utilized in the sphere of target monitoring detection. In some eventualities, a deep neural community is typically employed to implement a joint trace detection (tracking and object detection) community, the place a joint trace detection community refers to a community that is used to realize goal detection and iTag Pro goal hint together. In the existing joint monitoring detection network, the place and motion trail accuracy of the predicted moving target isn't excessive sufficient. The appliance supplies a goal tracking technique, a target tracking device and electronic tools, which can enhance the issues.

In one aspect, an embodiment of the current utility offers a goal tracking method, where the strategy includes: a first sub-network in a joint tracking detection network is used for extracting a first characteristic image from a target characteristic image, and a second sub-community in the joint monitoring detection community is used for extracting a second characteristic image from the goal characteristic picture, wherein the target characteristic picture is extracted from a video frame of a target video; fusing the second feature map extracted by the second sub-network to the primary feature map to acquire a fused feature map corresponding to the primary sub-community; buying first prediction info output by a primary sub-network in accordance with the fusion feature map, and buying second prediction data output by a second sub-community; primarily based on the primary prediction information and the second prediction data, figuring out the present position and the motion path of the moving target in the goal video.

Optionally, in the method supplied by the embodiment of the current application, the first subnetwork is a classification subnetwork, and the second subnetwork is a regression subnetwork or a tracking subnetwork. Optionally, in the tactic supplied by the embodiment of the present application, the first subnetwork is a regression subnetwork, and the second subnetwork is a classification subnetwork or a tracking subnetwork. In one other facet, an embodiment of the current software supplies a goal tracking apparatus, including: the system comprises a function acquisition module, iTag Pro a feature fusion module, a prediction module and a monitoring module. The feature acquisition module is used for detecting a first sub-community in the network by means of joint monitoring, extracting a primary feature map from a target feature map, and extracting a second characteristic map from the target function map via a second sub-community in the community via joint tracking, whereby the goal characteristic map is a function map extracted from a video body of a target video.