python keypoints

test.py

import norfair
import numpy as np
import time
from random import seed
from random import random
detection_threshold = 0.2
keypoint_dist_threshold = 1080 / 30
seed(1)


def keypoints_distance(detected_pose, tracked_pose):
    distances = np.linalg.norm(detected_pose.points - tracked_pose.estimate, axis=1)
    match_num = np.count_nonzero( (distances < keypoint_dist_threshold)
        * (detected_pose.scores > detection_threshold)
        * (tracked_pose.last_detection.scores > detection_threshold)
    )
    return 1 / (1 + match_num)

tracker = norfair.Tracker(distance_function=keypoints_distance, distance_threshold=0.3, detection_threshold=0.2)


def pose_format(keypoints,width,height):   # convert keypoints into required format
    convertkp=[]
    pose_score = []
    for kp in keypoints:
        if (kp[1]!=0 or kp[2]!=0):
         convertkp.append([kp[2]*width,kp[1]*height])
         pose_score.append(1.0)
        else:
         convertkp.append([0,0])
         pose_score.append(0.0)

    convertkp = np.array(convertkp, dtype=np.float32)
    pose_score = np.array(pose_score, dtype=np.float32)
    pose_results = []

    xmin = np.nanmin( convertkp[:, 0][convertkp[:, 0] > 0])
    xmax = np.nanmax(convertkp[:, 0])
    ymin = np.nanmin(convertkp[:, 1][convertkp[:, 1] > 0])
    ymax = np.nanmax(convertkp[:, 1])
    area = (xmax - xmin) * (ymax - ymin)
    pose_results.append({'keypoints': convertkp[:, :3],'score': 0.8, 'area': area, 'pose_score': pose_score, })
    return pose_results


def fakekeypoints():   # generate keypoints
    time.sleep(0.05)    # fake inference time
    # fake 18 keypoints for one person
    kp=[(0, 0, 0), (1, 0, 0), (2, 0, 0), (3, 0.3575015366077423, 0.27122947573661804),
     (4, 0.35617896914482117, 0.2848164737224579), (5, 0.38864418864250183, 0.2681225538253784),
     (6, 0.3827485740184784, 0.2904510200023651), (7, 0.4303573668003082, 0.26537251472473145),
     (8, 0.41974005103111267, 0.3006589412689209), (9, 0.45867863297462463, 0.27195972204208374),
     (10, 0.4473567605018616, 0.3024648427963257), (11, 0.46456748247146606, 0.2804908752441406),
     (12, 0.4614618420600891, 0.2982827126979828), (13, 0.5258688926696777, 0.29282426834106445),
     (14, 0.5186691284179688, 0.31580889225006104), (15, 0.5790600776672363, 0.3047087490558624),
     (16, 0.573983907699585, 0.3295579254627228), (17, 0.3861940801143646, 0.2790219783782959)]

    idx=0
    for _ in kp:        # add some noise
        kp[idx]=(kp[idx][0], kp[idx][1]+(random()/10), kp[idx][2]+(random()/10))
        idx=idx+1

    width=1920
    height=1080
    results=pose_format(kp,width,height)
    return results


prev_frame_time = 0
new_frame_time = 0
enable_tracker=False

if __name__ == '__main__':
    while(1):
        results=fakekeypoints()
        if enable_tracker:
            keypoint_dist_threshold = 1080 / 30  #img.shape[0] / 30
            detections = [norfair.Detection(p['keypoints'][:,0:2], scores=p['pose_score']) for p in results ]
            tracked_objects = tracker.update(detections=detections)
            #norfair.draw_tracked_objects(image, tracked_objects)

        new_frame_time = time.time()
        fps = 1 / (new_frame_time - prev_frame_time)
        prev_frame_time = new_frame_time
        print("fps is ", fps)