evaluation_example.py

import cv2
import numpy as np
from radar_optical_flow import RadarOpticalFlowExtrapolation, OpticalFlowMethod


def create_evaluation_test_images():
    img1 = np.zeros((400, 400, 3), dtype=np.uint8)
    cv2.circle(img1, (150, 150), 50, (0, 255, 0), -1)
    cv2.circle(img1, (250, 250), 30, (0, 200, 0), -1)
    
    img2 = np.zeros((400, 400, 3), dtype=np.uint8)
    cv2.circle(img2, (170, 170), 50, (0, 255, 0), -1)
    cv2.circle(img2, (270, 270), 30, (0, 200, 0), -1)
    
    img3 = np.zeros((400, 400, 3), dtype=np.uint8)
    cv2.circle(img3, (190, 190), 50, (0, 255, 0), -1)
    cv2.circle(img3, (290, 290), 30, (0, 200, 0), -1)
    
    cv2.imwrite('eval_img1.png', img1)
    cv2.imwrite('eval_img2.png', img2)
    cv2.imwrite('eval_img3.png', img3)
    
    return img1, img2, img3


def evaluate_all_methods():
    print("=== 雷达外推评估测试 ===\n")
    create_evaluation_test_images()
    # 使用真实的radar_frame系列图像
    img1 = cv2.imread('radar_frame_0.png')
    img2 = cv2.imread('radar_frame_1.png')
    img3 = cv2.imread('radar_frame_2.png')
    
    if img1 is None or img2 is None or img3 is None:
        create_evaluation_test_images()
      
    methods = ['farneback', 'lucas_kanade', 'dis']
    
    for method in methods:
        print(f"评估方法: {method}")
        print("-" * 40)
        
        extrapolator = RadarOpticalFlowExtrapolation(method=OpticalFlowMethod[method.upper()])
        
        predicted = extrapolator.extrapolate(img1, img2, steps=1)
        
        metrics = extrapolator.evaluate(predicted, img3, threshold=50)
        
        print(f"TS评分: {metrics['TS']:.3f}")
        print(f"POD探测率: {metrics['POD']:.3f}")
        print(f"FAR虚警率: {metrics['FAR']:.3f}")
        print(f"BIAS偏差: {metrics['BIAS']:.3f}")
        print(f"RMSE误差: {metrics['RMSE']:.1f}")
        print(f"命中数: {metrics['hits']}")
        print(f"漏报数: {metrics['misses']}")
        print(f"虚警数: {metrics['false_alarms']}")
        print()
        
        cv2.imwrite(f'eval_pred_{method}_radar.png', predicted)
    
    print("=== 指标说明 ===")
    print("TS评分: 0-1，越高越好，综合准确率")
    print("POD探测率: 0-1，越高越好，正确探测的比例")
    print("FAR虚警率: 0-1，越低越好，错误预报的比例")
    print("BIAS偏差: >1高估，<1低估，=1完美")
    print("RMSE误差: 越低越好，像素强度差异")


if __name__ == "__main__":
    evaluate_all_methods()