Stereo camera system RoboVision 2s Object detection package
Upper left:Actual image and detection result.The detected object is displayed in red frame.
Bottom left: Stereo camera Displays image converted from image data taken with RoboVision 2s to height information from the ground.
GUIでの設定によりカメラからの距離の情報に変換したイメージも表示可能。
Right: A view in which the result of object detection is displayed at the coordinates (camera position is the origin) viewed from above. Height information of the detected object is displayed in color.
Overview · Features
In addition to the distance information calculated from the parallax image in the stereo camera system RoboVison2s,
Visualize the size (width, height, depth) of the detected object and the position coordinates of the object from the camera in real time.
Setting of the object detection range and setting of the size to be detected can also be easily set on the GUI.
Detected video
Detection result integration screen
By using this system, it is also possible to detect objects with a height of about 10 cm, and it is possible to accurately detect the small objects on the road and the steps.
As for the object detection result, it is possible to output position and size information as well as the ID of the object.
Distance measurement realizes accuracy of less than 1% error at distance of 130 m if conditions are aligned. (Our measurement)
The object can be detected with high accuracy
From point cloud data obtained by parallax image of RoboVision 2s made by ZMP
Detection of overlapping in the depth direction can also detect objects when objects move in an overlapping manner.
Ability to detect objects accurately
The above image is the capture of the image of object detection.
Pedestrians start to overlap from image 2, and in pictures 3 to 5, the pedestrian behind the pedestrian starts to appear, the detection area increases little by little.
Since the object detection state is maintained from the point cloud data obtained from the gap of the bicycle, only the foot of the pedestrian hidden in the image 4 is detected.
In this way, since distance and height can be accurately measured for each image frame, it can be seen that good detection accuracy is obtained for object overlap.
It is possible to detect from the height of an object of about 10 cm
Since it can measure from a height of about 10 cm, if you move the mouse cursor to the data view that also acquired small objects on the road surface and small step measurements such as curb stones, the height and distance information of that point will be displayed on the GUI I will.
It is possible to detect from the height of an object of about 10 cm
■ Left: An example in which a small object of about 10 cm to 15 cm square was placed on the road and measured. Recognition as object, height information from road surface is displayed visually.
The road surface of 0 m is displayed in white, and the color changes as it increases. It is understood that the height information and the object detection are accurate.
■ Right: An example of taking a step, such as a curb, and measuring its height. The level difference of the curb of about 10 cm is clearly measured.
Realize highly accurate distance detection result
Under specific conditions, measure the distance of the object that is about 130 m away with an error within about 1%. It is possible to measure the distance of the object ahead 100 m or more.
Realize highly accurate distance detection result
■ Left: Raw image taken with RoboVision 2. The red spot is the distance from each point of about 40 m to 135 m measured by the laser range finder
Right: An image in which the distance information is calculated by extracting the edge portion of the building having a large change from the background
■ Left: Plot the distance measured by the laser distance measuring device and the measured distance at RoboVision 2
■ Right: The difference between the actual distance and the measurement of RoboVision 2. Even in the vicinity of 140 m, measurement error of 0.5 m or less was obtained.
※ Measurement results may vary depending on shooting environment and conditions. Object detection software will be provided only on Linux OS.
