US10719940B2 - Target Tracking Method And Device Oriented To Airborne-Based Monitoring Scenarios - Google Patents
Target detecting and tracking are two of the core duties in the sphere of visible surveillance. Relu activated fully-linked layers to derive an output of 4-dimensional bounding field data by regression, whereby the 4-dimensional bounding box information includes: horizontal coordinates of an higher left corner of the first rectangular bounding field, vertical coordinates of the upper left corner of the primary rectangular bounding box, a length of the primary rectangular bounding box, and a width of the first rectangular bounding field. FIG. Three is a structural diagram illustrating a goal tracking device oriented to airborne-based monitoring situations in accordance with an exemplary embodiment of the current disclosure. FIG. 4 is a structural diagram illustrating one other goal tracking device oriented to airborne-primarily based monitoring eventualities in accordance with an exemplary embodiment of the present disclosure. FIG. 1 is a flowchart diagram illustrating a target tracking method oriented to airborne-primarily based monitoring scenarios in line with an exemplary embodiment of the present disclosure. Step one hundred and one acquiring a video to-be-tracked of the target object in real time, and performing body decoding to the video to-be-tracked to extract a first body and a second body.
Step 102 trimming and capturing the primary frame to derive an image for first interest region, and trimming and capturing the second body to derive an image for best bluetooth tracker goal template and a picture for second curiosity area. N occasions that of a size and width data of the second rectangular bounding field, respectively. N could also be 2, that's, the length and width knowledge of the third rectangular bounding field are 2 occasions that of the size and width knowledge of the primary rectangular bounding field, respectively. 2 times that of the original information, obtaining a bounding box with an space four instances that of the original knowledge. In line with the smoothness assumption of motions, it is believed that the position of the goal object in the first body should be discovered in the interest area that the area has been expanded. Step 103 inputting the picture for goal template and the picture for first interest area right into a preset look tracker network to derive an look monitoring position.
Relu, and the number of channels for outputting the function map is 6, 12, 24, 36, 48, and sixty four in sequence. 3 for the remainder. To make sure the integrity of the spatial place info in the function map, the convolutional community doesn't embody any down-sampling pooling layer. Feature maps derived from different convolutional layers within the parallel two streams of the twin networks are cascaded and built-in utilizing the hierarchical function pyramid of the convolutional neural community whereas the convolution deepens repeatedly, respectively. This kernel is used for performing a cross-correlation calculation for dense sampling with sliding window type on the feature map, which is derived by cascading and integrating one stream corresponding to the image for first curiosity region, and a response map for appearance similarity can be derived. It may be seen that in the appearance tracker network, the tracking is in essence about deriving the position the place the target is positioned by a multi-scale dense sliding window search in the curiosity region.
The search is calculated based on the goal appearance similarity, that is, the appearance similarity between the goal template and the image of the searched place is calculated at every sliding window place. The place the place the similarity response is massive is very probably the position where the target is located. Step 104 inputting the image for first curiosity area and the image for second interest area right into a preset movement tracker network to derive a motion tracking position. Spotlight filter body distinction module, a foreground enhancing and background suppressing module in sequence, whereby each module is constructed based mostly on a convolutional neural network structure. Relu activated convolutional layers. Each of the number of outputted characteristic maps channel is three, wherein the function map is the distinction map for the enter image derived from the calculations. Spotlight filter body difference module to acquire a frame difference movement response map corresponding to the curiosity regions of two frames comprising previous body and subsequent body.
This multi-scale convolution design which is derived by cascading and secondary integrating three convolutional layers with totally different kernel sizes, goals to filter the movement noises attributable to the lens motions. Step 105 inputting the looks monitoring place and the motion tracking position into a deep integration network to derive an integrated closing tracking place. 1 convolution kernel to restore the output channel to a single channel, thereby teachably integrating the monitoring results to derive the ultimate monitoring place response map. Relu activated absolutely-linked layers, and a four-dimensional bounding field information is derived by regression for outputting. This embodiment combines two streams best bluetooth tracker networks in parallel within the strategy of monitoring the goal object, wherein the goal object's appearance and movement data are used to perform the positioning and tracking for the goal object, and the final tracking place is derived by integrating two occasions positioning information. FIG. 2 is a flowchart diagram illustrating a target monitoring method oriented to airborne-based monitoring situations in accordance to another exemplary embodiment of the current disclosure.