Eth data set download and related problems

ETH Dataset Download

Download address

Existing problems

I try to visualize the data , Because the data given by the data set is the data in the world coordinate system , We hope to show it in the picture , That is, to convert the world coordinate system to the pixel coordinate system . according to H.txt But after the transformation of the movie star matrix given in , The following problems were found
The code is as follows ：

class A:
    def __init__(self):
        self.h11 = 1.1048200e-02
        self.h12 = 6.6958900e-04
        self.h13 = -3.3295300e+00
        self.h21 = -1.5966000e-03
        self.h22 = 1.1632400e-02
        self.h23 = -5.3951400e+00
        self.h31 = 1.1190700e-04
        self.h32 = 1.3617400e-05
        self.h33 = 5.4276600e-01
        self.P = np.array([
            [self.h11,self.h12,self.h13],
            [self.h21,self.h22,self.h23],
            [self.h31,self.h32,self.h33]
        ])
        self.P1 = inv(self.P)

    def transform_coordinate(self, point):
        #  Input pixels , Back to world coordinates 
        x, y = point
        x_ = (self.h11*x + self.h12*y + self.h13)/(self.h31*x + self.h32*y + self.h33)
        y_ = (self.h21*x + self.h22*y + self.h23)/(self.h31*x + self.h32*y + self.h33)
        return (x_,y_)

    def w_to_p(self,point):
        point_ = np.append(point,1)
        p_point_ = np.matmul(self.P1,point_)
        p_point_ /= p_point_[-1]
        return p_point_[:-1]

among transform_coordinate,w_to_p Methods are pixel coordinates to world coordinates and world coordinates to pixel coordinates .

We execute the following visualization code , take obsmat.txt Visualize the first frame data in

	file_path = r"./ewap_dataset/seq_hotel/obsmat.txt"
    data = pd.read_csv(file_path,header=None,sep='\s+',names=["frame_number", "human_id", "pos_x", "pos_z", "pos_y", "v_x", "v_z", "v_y" ])[["frame_number", "human_id", "pos_x", "pos_y"]]
    data1 = data[data["frame_number"]==1]
    a = A()
    w_ps = []
    for i,row in data1.iterrows():
        point = np.array([row["pos_x"], row["pos_y"]])
        w_p = a.w_to_p(point)
        w_ps.append(w_p)
    
    import cv2

    img_path = r"./ewap_dataset/seq_hotel/jpg/1.jpg"
    img = cv2.imread(img_path)  #  Read the picture 
    for i in w_ps:
        cv2.circle(img, (int(i[0]), int(i[1])), 10, (0, 0, 255))

The output picture is as follows ：

It doesn't feel right …… But the code has been read many times without any problem .
According to the advice of elder martial sister, I found that x,y Change the position to output correctly

	file_path = r"./ewap_dataset/seq_hotel/obsmat.txt"
    data = pd.read_csv(file_path,header=None,sep='\s+',names=["frame_number", "human_id", "pos_x", "pos_z", "pos_y", "v_x", "v_z", "v_y" ])[["frame_number", "human_id", "pos_x", "pos_y"]]
    data1 = data[data["frame_number"]==1]
    a = A()
    w_ps = []
    for i,row in data1.iterrows():
        point = np.array([row["pos_x"], row["pos_y"]])
        w_p = a.w_to_p(point)
        w_ps.append(w_p)
    
    import cv2

    img_path = r"./ewap_dataset/seq_hotel/jpg/1.jpg"
    img = cv2.imread(img_path)  #  Read the picture 
    for i in w_ps:
        cv2.circle(img, (int(i[1]), int(i[0])), 10, (0, 0, 255))

Then the data here should be wrong , Or the homography matrix is wrong . Personal feeling , If you know the boss, please correct ！！