pos_data.append(pos_tem) feature_data.append(feature_particle) patch_data.append(patch_particle)#存放每一帧patch的34*32HOG count = j-1#有效粒子计数 ''' 计算每个有效patch的权重w ''' res =[] res.append(feature_particle) feature_2=np.array(res) print(feature_2.shape) tem_w=cnc_rnn.predict(feature_2) print tem_w w = 1-((1-tem_w[0][0])**2+(0- tem_w[0][1])**2)**(0.5) #gc.collect() print i,j print w #print (w) w_data.append(w)#存放每一帧patch的权重w(未归一化) wx = pos_tem[0]*w#每一个patch的权重w 乘以 对应patch的x坐标 wy = pos_tem[1]*w#每一个patch的权重w 乘以 对应patch的y坐标 wxim.append(wx)#存放当前im的所有patch带权坐标x wyim.append(wy)#存放当前im的所有patch带权坐标y iniw[count] = iniw[count] + w#这里权值向量iniw在无效粒子处权值w为0
import hog import cv2 import numpy as np import sample raw = 16 col = 32 img_dir = './balltrain/test.jpg' image = cv2.imread(img_dir, cv2.IMREAD_GRAYSCALE) feature = sample.HOG_feature(image) res = [] res.append(feature) feature_2 = np.array(res) print(image.shape) print(feature_2.shape) ''' example: feature_2=[ [image] [image] [image] ] [image]=[ [+++++] [+++++] [+++++] ] feature_2.shape=(1,9,32) ''' import cnc_rnn prediction = cnc_rnn.predict(feature_2) print(prediction) print(prediction[0][0])
except: print(i) print(j) pass #计算当前im期望位置prex prey,归一化权重 data = np.array(feature_data) print(data.shape) print(i) print(len(feature_particle)) try: tem_w=cnc_rnn.predict(data) except: print(data) print(data.shape) print(type(feature_particle)) for each_one in data: print(type(each_one)) break #break #for each_one in tem_w: #print(each_one) ''' wsum = 0