A detection and tracking method and system based on circular frame buffer

Abstract

The present invention relates to a method for detection and tracking based on buffering, including several detectors, including the following steps: A, establishing a circular buffer on the time axis; B, extracting a video according to a strategy from a continuous frame sequence input in real time Frame and record the frame number; C. Send the selected frames to several target detectors for parallel target detection; D. Address the buffer according to the detected source frame number and read the corresponding source frame number The frame is set as the start frame, and the start frame is StartFrame(a), StartFrame(b), StartFrame(c), StartFrame(d)....etc; E. For each detected target in the buffer and the corresponding start frame for initialization processing and tracking processing for the target. The proportion of time that can be used for actual calculation in the tracking process is increased, which improves the tracking performance and supports a larger tracking target set.

Description

technical field [0001] The invention belongs to the technical field of detection and tracking of frame data, and in particular relates to a detection and tracking method and system based on a circular frame buffer. Background technique [0002] At present, based on video intelligent analysis and processing, it has been more and more widely used in various fields. In these applications, a key issue is: how to efficiently detect objects of interest from continuous multi-frame video and / or images, and continuously track these detected objects; that is, the objects in the video stream detection and tracking issues. [0003] Typically, a general solution to this type of problem is framed as figure 1 shown. [0004] In the above detection and tracking processing framework, usually the algorithms used for target detection and tracking consume a lot of computing resources, and it is usually impossible to realize real-time data processing in a single thread of a single processor. ...

AI Technical Summary

Problems solved by technology

In these applications, a key issue is: how to efficiently detect objects of interest from continuous multi-frame video and / or images, and continuously track these detected objects; that is, the objects in the video stream detection and tracking of

[0005] However, the commonly used or existing solutions mentioned above did not solve a key bottleneck problem, resulting in a greatly limited parallelism, for the following reasons:

[0006] 1. Although the frame-by-frame tracking calculation can be performed in a relatively high concurrency, as mentioned above, since the target of the frame-by-frame tracking calculation comes from the target detection process, when it is in the detection frame, the tracking calculation will be blocked. Do not proceed until the object detection part has finished detecting and maintaining the set of tracked objects

2. Since the target detection operation is time-consuming, the blockage tracking will run for a long time

3. At the same time, since the detection operation and the maintenance of the tracking target set are serial, the target detection operation cannot be parallelized

The entire processing framework cannot run in parallel at full speed, and the speed is limited by the serial process of target detection

4. At the same time, due to the serial operation of target detection, it is necessary to introduce frame extraction to meet the real-time processing requirements, which greatly increases the probability of target missed detection

[0011] 1: The problem of target loss and omission caused by frame detection due to insufficient parallel detection

[0012] 2: Since the blocking of detection compresses the available computing time for tracking, the processing speed / scale of target tracking is limited.

[0013] As mentioned above, in the technical solution, there are some key bottleneck problems, which lead to the parallelism of the overall operation and limitation, and the limitation of parallelism will cause system related problems such as:

[0014] 1: At the same time, due to the serial operation of target detection, it is necessary to introduce frame extraction to meet the real-time processing requirements, which greatly increases the probability of target missed detection

[0015] 2: Since the blocking of detection compresses the available computing time for tracking, the processing speed / scale of target tracking is limited.

[0016] 3: Detection target failure due to detection delay

Images