A video stream generation system and method based on multi-model superposition

Abstract

The present invention discloses a video stream generation system and method based on multi-model superposition, which deploys multiple target detection models in parallel, processes the image frames of the video stream at the same time, and then aggregates and processes the detection results of each model to generate a new target visualization video stream. The present invention can meet the needs of real-time target detection of video streams in a variety of complex scenarios. After the detection of each model is completed, the system can aggregate and process the results and generate a new target visualization video stream, thereby enhancing the adaptability and practicality of the detection system.

Description

Technical Field

[0001] The present invention relates to a system and method for processing video streams in parallel by using multiple target detection models, and aggregating and processing the results to generate a new target visualization video stream, belonging to the technical field of video stream processing and target detection. Background Art

[0002] With the rapid development of video surveillance, autonomous driving, intelligent security and other fields, the accuracy and real-time requirements for target detection in video streams are becoming increasingly higher. Traditional single-model target detection methods often have problems such as low detection efficiency and long processing time when facing complex and diverse target types and large amounts of video data, making it difficult to meet the real-time requirements in actual application scenarios. Summary of the invention

[0003] The technical problem to be solved by the present invention is that the traditional single-model target detection method often has problems such as low detection efficiency and long processing time when faced with complex and diverse target types and large amounts of video data, and it is difficult to meet the real-time requirements in actual application scenarios.

[0004] In order to solve the above technical problems, the present invention discloses a video stream generation system based on multi-model superposition, which is characterized by comprising:

[0005] A video frame acquisition module, used to acquire video frames from an input video stream at a predetermined frame rate;

[0006] The model interface deployment module is used to pre-deploy the model interfaces of multiple target detection models.

[0007] The parallel detection module is used to distribute the collected video frames to multiple pre-deployed model interfaces at the same time, so that multiple target detection models can perform detection tasks in parallel;

[0008] A result summary module is used to receive the detection results of multiple target detection models and organize and summarize them;

[0009] A visualization module is used to visualize the aggregated detection results output by the result aggregation module on the original video frame to generate a visualization result frame of multi-model detection;

[0010] The video stream merging module is used to merge all visualization result frames into a new video stream for output.

[0011] Preferably, all of the target detection models target different target types or have different detection characteristics to adapt to diverse detection requirements.

[0012] Preferably, the visualization module marks the target position and category detected by each target detection model on the visualization result frame.

[0013] Preferably, the video stream merging module merges all the visualization result frames into a new video stream according to the time sequence of the video.

[0014] Another technical solution of the present invention is to disclose a method for generating a video stream based on multi-model superposition, which adopts the above-mentioned video stream generation system based on multi-model superposition, and is characterized in that it includes the following steps:

[0015] Step 1: Start the video frame acquisition module to acquire video frames from the video stream according to a predetermined frame rate;

[0016] Step 2: Send the collected video frames to multiple pre-deployed model interfaces at the same time to trigger parallel detection tasks;

[0017] Step 3: Multiple target detection models perform target detection on video frames in parallel, and each outputs a detection result;

[0018] Step 4: The result summary module collects and summarizes the detection results of each target detection model;

[0019] Step 5: The visualization module performs visualization processing on the original video frame according to the summary results output by the result summary module to generate a visualization result frame;

[0020] Step 6: The video stream merging module merges the visualization result frames into a new video stream output.

[0021] The present invention discloses a system and method for multi-target detection and visualization of video streams based on multi-model parallel processing, which deploys multiple target detection models in parallel, processes image frames of video streams at the same time, and then aggregates and processes the detection results of each model to generate a new target visualization video stream. The present invention can meet the needs of real-time target detection of video streams in a variety of complex scenarios. After the detection of each model is completed, the system can aggregate and process the results and generate a new target visualization video stream, thereby enhancing the adaptability and practicality of the detection system. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 It is the system flow chart;

[0023] Figure 2 This is a diagram of the video frame processing module. DETAILED DESCRIPTION

[0024] The present invention will be further described below in conjunction with specific embodiments. It should be understood that these embodiments are only used to illustrate the present invention and are not intended to limit the scope of the present invention. In addition, it should be understood that after reading the content taught by the present invention, those skilled in the art can make various changes or modifications to the present invention, and these equivalent forms fall within the scope limited by the appended claims of the application equally.

[0025] One aspect of an embodiment of the present invention is to disclose a video stream generation system based on multi-model superposition, including:

[0026] Video frame acquisition module: used to acquire video frames from the input video stream at a predetermined frame rate.

[0027] Model interface deployment module: pre-deploys model interfaces of multiple target detection models. These target detection models target different target types or have different detection characteristics to meet diverse detection needs.

[0028] Parallel detection module: The collected video frames are distributed to multiple pre-deployed model interfaces at the same time, so that multiple models can perform detection tasks in parallel, significantly reducing the detection time.

[0029] Result aggregation module: receives the detection results of multiple target detection models, and organizes and aggregates them for subsequent processing.

[0030] Visualization module: Visualize the aggregated detection results on the original video frame, mark the target location, category and other information detected by each target detection model, and generate a visualization result frame for multi-model detection.

[0031] Video stream merging module: merges all visualization result frames into a new video stream in the time order of the video for output. The output video stream fully presents the visualization results of multi-model multi-target detection in the video stream.

[0032] Another aspect of an embodiment of the present invention is to disclose a method for generating a video stream based on multi-model superposition, which specifically includes the following steps:

[0033] Step 1: Start the video frame acquisition module to acquire video frames from the video stream at a predetermined frame rate.

[0034] Step 2: Send the collected video frames to multiple pre-deployed model interfaces at the same time to trigger parallel detection tasks.

[0035] Step 3: Multiple target detection models perform target detection on video frames in parallel and output detection results respectively.

[0036] Step 4: The result summary module collects and summarizes the detection results of each model.

[0037] Step 5: The visualization module performs visualization processing on the original video frame based on the summary results to generate a visualization result frame.

[0038] Step 6: The video stream merging module merges the visualization result frames into a new video stream output.

[0039] A feasible implementation method of the above technical solution is: in the system initialization stage, firstly, the deployment of multiple model interfaces is completed through the model interface deployment module to ensure that each target detection model can normally receive video frame data and perform detection operations. When the video stream is input, the video frame acquisition module extracts video frames according to the set frame rate (such as 30 frames per second). The collected video frames are immediately distributed to each target detection model interface by the parallel detection module, and each target detection model starts detection at the same time. For example, target detection model 1 focuses on detecting pedestrians, target detection model 2 detects vehicles, and target detection model 3 detects specific objects. Each model runs independently without interfering with each other, making full use of computing resources, and greatly shortening the overall detection time. After the detection is completed, the result summary module integrates the detection results of each target detection model, such as the coordinate position of the target, the category to which it belongs, and other information. According to these summary results, the visualization module uses different marking methods (such as boxes of different colors, icons, etc.) on the original video frame to mark the targets detected by each target detection model and generate visualization result frames. Finally, the video stream merging module splices these visualization result frames into a new video stream in the correct time sequence, which can be used for real-time monitoring display, storage or further analysis and processing.

Claims

1. A video stream generation system based on multi-model superposition, characterized in that: include: A video frame acquisition module, used to acquire video frames from an input video stream at a predetermined frame rate; The model interface deployment module is used to pre-deploy the model interfaces of multiple target detection models. The parallel detection module is used to distribute the collected video frames to multiple pre-deployed model interfaces at the same time, so that multiple target detection models can perform detection tasks in parallel; A result summary module is used to receive the detection results of multiple target detection models and organize and summarize them; A visualization module is used to visualize the aggregated detection results output by the result aggregation module on the original video frame to generate a visualization result frame of multi-model detection; The video stream merging module is used to merge all visualization result frames into a new video stream for output.

2. A video stream generation system based on multi-model superposition as claimed in claim 1, characterized in that: All of the target detection models are targeted at different target types or have different detection characteristics to adapt to diverse detection requirements.

3. A video stream generation system based on multi-model superposition as claimed in claim 1, characterized in that: The visualization module marks the target positions and categories detected by each target detection model on the visualization result frame.

4. A video stream generation system based on multi-model superposition as claimed in claim 1, characterized in that: The video stream merging module merges all the visualization result frames into a new video stream according to the time sequence of the video.

5. A method for generating a video stream based on multi-model superposition, using the video stream generation system based on multi-model superposition according to claim 1, characterized in that: The following steps are involved: Step 1: Start the video frame acquisition module to acquire video frames from the video stream according to a predetermined frame rate; Step 2: Send the collected video frames to multiple pre-deployed model interfaces at the same time to trigger parallel detection tasks; Step 3: Multiple target detection models perform target detection on video frames in parallel, and each outputs a detection result; Step 4: The result summary module collects and summarizes the detection results of each target detection model; Step 5: The visualization module performs visualization processing on the original video frame according to the summary results output by the result summary module to generate a visualization result frame; Step 6: The video stream merging module merges the visualization result frames into a new video stream output.

Images