Direction independent speed detection based on distance estimation with fixed focus camera

Abstract

The invention relates to a method for direction independent speed detection using vehicle (2) distance estimation from fixed focus cameras (1).

Description

[0001] DIRECTION INDEPENDENT SPEED DETECTION BASED ON DISTANCE ESTIMATION WITH FIXED FOCUS CAMERA

[0002] Technical Field

[0003] The invention relates to a method for direction independent speed detection using vehicle distance estimation from fixed focus cameras.

[0004] Prior Art

[0005] Radar systems are most commonly used for speed detection. However, these systems generally detect the speed of a single vehicle and can be more costly than camera systems. In addition, they cannot be applied everywhere and they need an operator to use the system.

[0006] JP2005323021A discloses a in-vehicle imaging system and imaging method.

[0007] CN 107886523A discloses a vehicle target movement speed detection method based on UAV multi-source images.

[0008] When the existing studies in the prior art are examined, it is necessary to develop a method that enables the realisation of direction independent speed detection with vehicle distance estimation from fixed focus cameras.

[0009] Objectives of the Invention

[0010] The object of the present invention is to provide a method for performing direction independent speed detection with vehicle distance estimation from fixed focus cameras.

[0011] Another object of the present invention is to provide a method for speed detection in an urban environment, e.g. on streets, etc., using simple fixed-focus cameras (Urban Security Management System cameras). Detailed Description of the Invention

[0012] An exemplary embodiment of a speed detection method for achieving the objects of the present invention is shown in the attached figures.

[0013] Figures;

[0014] Figure 1: A representative representation of the displacement vectors of the vehicle in the camera in the speed detection method of the invention.

[0015] Figure 2: A representative representation of the calculation of the three- dimensional position of an object having distance information relative to a camera in the speed detection method of the invention.

[0016] The parts in the figures are numbered individually, and the corresponding descriptions are given below.

[0017] 1. Camera

[0018] 2. Vehicle vl: Camera-vehicle position vector at initial position v2: Camera-vehicle position vector at the second position

[0019] Vn: Position vector of the object at a given point

[0020] Ix: X-axis coordinate of the pixel position of the object on the image

[0021] Iy: Y-axis coordinate of the pixel position of the object on the image

[0022] Cx: X-axis coordinate of the image centre

[0023] Cy: Y-axis coordinate of the image centre f: Focal length d: Estimated distance

[0024] X: X-axis

[0025] Y: Y-axis

[0026] Z: Z-axis t: Duration

[0027] A method for direction independent speed detection using vehicle (2) distance estimation from fixed focus cameras (1), comprising the steps of; taking the image of the vehicle (2) with camera (1), finding the 3D wireframes and distances to the camera (1) of the vehicles (2) imaged by the camera (1) with the monocular 3D object detection deep learning method1, calculating the position of the object in space relative to the camera (1) from the images of the vehicles (2) taken during a given duration t, using the external parameter of the camera (1) and the distance of the object at any pixel of the camera (1) from the camera (1) using geometric calculation with a pinhole camera model, calculating the difference between calculated two separate locations, calculating the speed by dividing the calculated difference by the duration.

[0028] In the inventive method, the vehicle (2) is first imaged with a fixed focus camera (1). The 3D wireframes of the vehicles (2) imaged with camera (1) and their distances to camera (1) are found by monocular 3D object detection deep learning method.1In this way, the camera (1) - vehicle (2) vector (vl) at the initial position is calculated. (Figure 1)

[0029] Then, for vehicle (2) whose vector vl is known, vehicle tracking is performed in camera (1) for a certain duration (t) during N frames. The 3D wireframes of the vehicles (2) imaged with camera (1) and their distances to camera (1) are found by monocular 3D object detection deep learning method.1In this way, the camera (1)

[0030] - vehicle (2) vector (v2) at the second position is calculated (Figure 1).

[0031] In the inventive method, monocular 3D object detection is performed by GLENet: Generative label uncertainty, estimation method for enhancing 3D object detectors2or 3D object detection methods (KITTI etc.) can be used.

[0032] Then, from the images of the vehicles (2) taken during a certain duration (t), the position of the object in space relative to the camera (1) is geometrically calculated with the pinhole camera model using the information of the external parameter of the camera (1) and the distance of the object to the camera (1) at any pixel of the camera (1). (Figure 2)

[0033] When calculating the position of the object in space, the coordinates of the object with respect to the focal point are calculated with the help of Formula I, where Cxand Cyare the image centre, f is the focal length, Ixand Iyare the pixel position of the object on the image, and d is the estimated distance, all units being in metres. Vn is the position vector of the object at a certain point.

[0034] The difference between the two different positions is calculated. The difference between the two calculated positions is divided by the duration (t) to determine the speed.

[0035] The invention enables the detection of vehicle (2) speeds with a simple monocular camera (1) in traffic systems. The method of the invention can be used by police units in traffic areas, as well as by security units in campus-style areas to determine whether vehicle (2) speed limits have been exceeded. References:

[0036] [1] Eskil Jorgensen, Christopher Zach, Fredrik Kahl, “Monocular 3D Object Detection and Box Fitting Trained End-to-End Using Intersection-over-Union Loss” (arXiv: 1906.08070) [2] Yifan Zhang, Qijian Zhang, Zhiyu Zhu, Junhui Hou, Yixuan Yuan, “GLENet:

[0037] Boosting 3D Object Detectors with Generative Label Uncertainty Estimation” (arXiv:2207.02466)

Claims

CLAIMS1. A method for direction independent speed detection using vehicle (2) distance estimation from fixed focus cameras (1), characterized by, comprising the steps of; taking the image of the vehicle (2) with camera (1), finding the 3D wireframes and distances to the camera (1) of the vehicles (2) imaged by the camera (1) with the monocular 3D object detection deep learning method1, calculating the position of the object in space relative to the camera (1) from the images of the vehicles (2) taken during a given duration t, using the external parameter of the camera (1) and the distance of the object at any pixel of the camera (1) from the camera (1) using geometric calculation with a pinhole camera model, calculating the difference between calculated two separate locations, calculating the speed by dividing the calculated difference by the duration.

2. A method accoring to the claim 1, characterized by, instead of monocular 3D object detection deep learning method, GLENet: Generative label uncertainty, estimation method for enhancing 3D object detectors2or 3D object detection methods (KITTI etc.) can be used.

Images