Video Map combines AI target recognition technology, uses TensorFlow framework, carries out in-depth learning and Model Training on the target in the video, automatically identifies the type of the target from the video, and identifies the type and location of the target.
It provides 80 target types, such as cars, buses, trains, traffic lights, pedestrians, water cups, notebooks and other traffic targets and daily necessities. At the same time, it supports setting the style of different target identification boxes to facilitate the distinction of target types. Video object recognition can be applied to real-time monitoring of road traffic congestion, such as congestion of motor vehicle flow, non-motor vehicle flow and pedestrian flow.
The operation process of Object Detection function is divided into the following two steps:
Step 1: Detection Settings
Before Object Detection, you need to perform the Detection Settings operation to set parameters such as Detection Area, Detection Model, Detection Type, and Identification Style.
Function entrance
Video Analysis tab-> Traffic Analysis group-> Detection Settings button.
Parameter Description
Users can add detection models by customizing. Select Custom from the Model Settings Drop-down Button to pop up the Custom Model Management dialog box, where you can manage models by clicking the Add, Import, Modify, and Delete buttons. Click the Add button to pop up the Add Model dialog box. In the dialog box, you can complete the addition of the model by User-defined model algorithm, Model File, Type File, training Picture Size, Model Name, and model description parameters. Click the Import button to Import Model File (.sdm) from the Local File.
- Detection Area: Check this box, You can set the Detection Area by Select Dataset, Select Object, Draw Rectangle or Draw Polygon in Settings.. Drop-down Button. This checkbox is not checked by default.
- Model Settings: Support YOLOv5x, YOLOv5l, YOLOv5m, YOLOv5s, YOLOv5n, VisDrone, road damage, Smoke Fire, YOLOv7-E6E, YOLOv7-D6, YOLOv7-E6, YOLOv7-W6, YOLOv7-X, YOLOv7, YOLOv7-Tiny 15 Model Settings, default to YOLOv5x mode. VisDrone model is suitable for orthographic Video Detection, such as the orthographic video shot by UAV; other models are suitable for oblique Video Detection, such as the video shot by road camera. Please refer to the Video Analysis Environment Configuration for the model acquisition method. The specific description of the model is as follows: The largest model in The larger model in The medium size model in between accuracy and speed The smallest model in A variant of for Nano devices Model in Model with the second highest detection accuracy but the second slowest detection rate in Model with higher detection accuracy but slower detection rate Model with high detection accuracy but slow detection rate Model with low detection accuracy but fast detection rate Model with lower detection accuracy but faster detection rate Detection model Detection Model in parks and forests.
Model Features Applicable scenarios YOLOv5x the YOLOv5 series, where X is extra large, has the best detection accuracy but the slowest detection speed. It is applicable to videos with oblique viewing angles, and the specific model selection should be based on business requirements. When the service does not need to be detected every frame (detected every few seconds), the model with high detection accuracy is selected; when the service needs to be detected every frame, the model with high detection rate is selected.
For example, in farmland protection, when it is necessary to provide early warning for construction phenomena, it is not necessary to detect every frame (every few seconds), and the model with high detection accuracy can be selected; while in the statistics of traffic flow at intersections, it is necessary to detect every frame, and the model with high detection rate can be selected.
YOLOv5s is suitable for real-time Object Detection tasks on mobile devices or edge devices.
YOLOv5n provides accuracy for edge devices.
YOLOv7-Tiny is suitable for hardware configuration environment of edge GPU.
YOLOv5l the YOLOv5 series, where l represents large, has a relatively high detection accuracy, but the detection speed is slow. YOLOv5m the YOLOv5 series, where m stands for medium, provides a good balance YOLOv5s the YOLOv5 series, where s stands for small, has the fastest detection speed and the lowest detection accuracy YOLOv5n the YOLOv5 family optimized YOLOv7-E6E the YOLOv7 series that detects Highest Accuracy but detects the slowest rate YOLOv7-D6 the YOLOv7 series YOLOv7-E6 in YOLOv7 series YOLOv7-W6 in YOLOv7 series YOLOv7-X in YOLOv7 series YOLOv7 in YOLOv7 series YOLOv7-Tiny The YOLOv7 family detects Low Accuracy, but detects the fastest model VisDrone Orthographic perspective model, supporting the identification of UAV aerial photography: five Detection Types of pedestrians, cars, buses, vans and trucks. Traffic management, social activity flow monitoring The road is damaged for road cracks, potholes and other phenomena Suitable for road maintenance Smoke Fire for Fire Smoke It is suitable for fire monitoring - Detection Type: Add, delete, select all and reverse selection tools are provided in the toolbar of the Detection Type list, and the Detection Type can be set through these tools. At the same time, the settings of effective width and effective height are also provided in the Detection Type list. There are six default Detection Types: Bus, Bicycle, Motorcycle, Pedestrian, Truck and Automobile. The Effective width and height of Bus, Truck and Automobile are 74 by default. There is no setting for other types.
- Preview: Application provides a preview view to preview the Style Settings of the Detection Type.
- Style Settings: Style Settings is used to set the Display Effects of Detection Type. You can set the line width and border color. At the same time, whether Show Lable is checked is supported, and the font, size, bold and right italic settings of the label are provided. The default line width is 4; the border color is white except for the pedestrian; the Show Lable check box is not checked by default.
After configuring the environment for Video Analysis, you can activate the Object Detection button. See the Video AnalysisEnvironment configuration page for a detailed description of the environment configuration.
Step 2: Execute Object Detection
Function entrance
Video Analysis tab-> Traffic Analysis group-> Object Detection button.
Main operation
As shown in the figure below, click the Object Detection button to start Play Video from the beginning, detect the type of the object in the video, and dynamically identify the object in the video through the mark box.
Related topics