系列文章目錄

目標(biāo)跟蹤——SORT算法原理淺析
目標(biāo)跟蹤——Deep Sort算法原理淺析
基于yolov5與Deep Sort的流量統(tǒng)計(jì)與軌跡跟蹤

---

文章目錄

• 系列文章目錄
• 前言
• 一、整體目錄結(jié)構(gòu)
• 二、Deep Sort代碼參數(shù)解釋
• 三、代碼展示
• 總結(jié)

---

前言

先來(lái)看下實(shí)現(xiàn)效果：

上圖展示了用yolov5作為檢測(cè)器，Deep Sort為追蹤器實(shí)現(xiàn)了對(duì)車(chē)流量的統(tǒng)計(jì)并繪制了每輛車(chē)的運(yùn)行軌跡。

---

一、整體目錄結(jié)構(gòu)

下圖展示了項(xiàng)目的整體目錄結(jié)構(gòu)：

其中：
deep_sort文件下為目標(biāo)跟蹤相關(guān)代碼；
weights文件夾下存放yolov5檢測(cè)模型；
demo.py針對(duì)讀取的視頻進(jìn)行目標(biāo)追蹤
objdetector.py封裝的一個(gè)目標(biāo)檢測(cè)器，對(duì)視頻中的物體進(jìn)行檢測(cè)
objtracker.py封裝了一個(gè)目標(biāo)追蹤器，對(duì)檢測(cè)的物體進(jìn)行追蹤

二、Deep Sort代碼參數(shù)解釋

deep_sort/configs/deep_sort.yaml文件里保存了Deep Sort算法的配置參數(shù)：


這些參數(shù)依次的含義為：

REID_CKPT:特征提取權(quán)重的目錄路徑

MAX_DIST: 最大余弦距離，用于級(jí)聯(lián)匹配，如果大于該閾值，則忽略

MIN_CONFIDENCE:檢測(cè)結(jié)果置信度閾值

NMS_MAX_OVERLAP: 非極大抑制閾值，設(shè)置為1代表不進(jìn)行抑制

MAX_IOU_DISTANCE: 最大IOU閾值

MAX_AGE:最大壽命，也就是經(jīng)過(guò)MAX_AGE幀沒(méi)有追蹤到該物體，就將該軌跡變?yōu)閯h除態(tài)

N_INIT: 最高擊中次數(shù)，如果擊中該次數(shù)，就由不確定態(tài)轉(zhuǎn)為確定態(tài)

NN_BUDGET: 最大保存特征幀數(shù)，如果超過(guò)該幀數(shù)，將進(jìn)行滾動(dòng)保存

---

三、代碼展示

下面給出demo.py的代碼：

import numpy as npimport objtracker from objdetector import Detector import cv2VIDEO_PATH = './video/test_traffic.mp4'if __name__ == '__main__':# 根據(jù)視頻尺寸，填充供撞線計(jì)算使用的polygonwidth = 1920height = 1080mask_image_temp = np.zeros((height, width), dtype=np.uint8)# 用于記錄軌跡信息pts = {}# 填充第一個(gè)撞線polygon（藍(lán)色）list_pts_blue = [[204, 305], [227, 431], [605, 522], [1101, 464], [1900, 601], [1902, 495], [1125, 379], [604, 437],[299, 375], [267, 289]]ndarray_pts_blue = np.array(list_pts_blue, np.int32)polygon_blue_value_1 = cv2.fillPoly(mask_image_temp, [ndarray_pts_blue], color=1)polygon_blue_value_1 = polygon_blue_value_1[:, :, np.newaxis]# 填充第二個(gè)撞線polygon（黃色）mask_image_temp = np.zeros((height, width), dtype=np.uint8)list_pts_yellow = [[181, 305], [207, 442], [603, 544], [1107, 485], [1898, 625], [1893, 701], [1101, 568],[594, 637], [118, 483], [109, 303]]ndarray_pts_yellow = np.array(list_pts_yellow, np.int32)polygon_yellow_value_2 = cv2.fillPoly(mask_image_temp, [ndarray_pts_yellow], color=2)polygon_yellow_value_2 = polygon_yellow_value_2[:, :, np.newaxis]# 撞線檢測(cè)用的mask，包含2個(gè)polygon，（值范圍 0、1、2），供撞線計(jì)算使用polygon_mask_blue_and_yellow = polygon_blue_value_1 + polygon_yellow_value_2# 縮小尺寸，1920x1080->960x540polygon_mask_blue_and_yellow = cv2.resize(polygon_mask_blue_and_yellow, (width // 2, height // 2))# 藍(lán) 色盤(pán) b,g,rblue_color_plate = [255, 0, 0]# 藍(lán) polygon圖片blue_image = np.array(polygon_blue_value_1 * blue_color_plate, np.uint8)# 黃 色盤(pán)yellow_color_plate = [0, 255, 255]# 黃 polygon圖片yellow_image = np.array(polygon_yellow_value_2 * yellow_color_plate, np.uint8)# 彩色圖片（值范圍 0-255）color_polygons_image = blue_image + yellow_image# 縮小尺寸，1920x1080->960x540color_polygons_image = cv2.resize(color_polygons_image, (width // 2, height // 2))# list 與藍(lán)色polygon重疊list_overlapping_blue_polygon = []# list 與黃色polygon重疊list_overlapping_yellow_polygon = []# 下行數(shù)量down_count = 0# 上行數(shù)量up_count = 0font_draw_number = cv2.FONT_HERSHEY_SIMPLEXdraw_text_postion = (int((width / 2) * 0.01), int((height / 2) * 0.05))# 實(shí)例化yolov5檢測(cè)器detector = Detector()# 打開(kāi)視頻capture = cv2.VideoCapture(VIDEO_PATH)while True:# 讀取每幀圖片_, im = capture.read()if im is None:break# 縮小尺寸，1920x1080->960x540im = cv2.resize(im, (width // 2, height // 2))list_bboxs = []# 更新跟蹤器output_image_frame, list_bboxs = objtracker.update(detector, im)# 輸出圖片output_image_frame = cv2.add(output_image_frame, color_polygons_image)if len(list_bboxs) > 0:# ----------------------判斷撞線----------------------for item_bbox in list_bboxs:x1, y1, x2, y2, _, track_id = item_bbox# 撞線檢測(cè)點(diǎn)，(x1，y1)，y方向偏移比例 0.0~1.0y1_offset = int(y1 + ((y2 - y1) * 0.5))x1_offset = int(x1 + ((x2 - x1) * 0.5))# 撞線的點(diǎn)y = y1_offsetx = x1_offset# 然后每檢測(cè)出一個(gè)預(yù)測(cè)框，就將中心點(diǎn)加入隊(duì)列center = (x, y)if track_id in pts:pts[track_id].append(center)else:pts[track_id] = []pts[track_id].append(center)thickness = 2cv2.circle(output_image_frame, (center), 1, [255, 255, 255], thickness)for j in range(1, len(pts[track_id])):if pts[track_id][j - 1] is None or pts[track_id][j] is None:continuecv2.line(output_image_frame, (pts[track_id][j - 1]), (pts[track_id][j]), [255, 255, 255], thickness)if polygon_mask_blue_and_yellow[y, x] == 1:# 如果撞 藍(lán)polygonif track_id not in list_overlapping_blue_polygon:list_overlapping_blue_polygon.append(track_id)# 判斷 黃polygon list里是否有此 track_id# 有此track_id，則認(rèn)為是 UP (上行)方向if track_id in list_overlapping_yellow_polygon:# 上行+1up_count += 1print('up count:', up_count, ', up id:', list_overlapping_yellow_polygon)# 刪除 黃polygon list 中的此idlist_overlapping_yellow_polygon.remove(track_id)elif polygon_mask_blue_and_yellow[y, x] == 2:# 如果撞 黃polygonif track_id not in list_overlapping_yellow_polygon:list_overlapping_yellow_polygon.append(track_id)# 判斷 藍(lán)polygon list 里是否有此 track_id# 有此 track_id，則 認(rèn)為是 DOWN（下行）方向if track_id in list_overlapping_blue_polygon:# 下行+1down_count += 1print('down count:', down_count, ', down id:', list_overlapping_blue_polygon)# 刪除 藍(lán)polygon list 中的此idlist_overlapping_blue_polygon.remove(track_id)# ----------------------清除無(wú)用id----------------------list_overlapping_all = list_overlapping_yellow_polygon + list_overlapping_blue_polygonfor id1 in list_overlapping_all:is_found = Falsefor _, _, _, _, _, bbox_id in list_bboxs:if bbox_id == id1:is_found = Trueif not is_found:# 如果沒(méi)找到，刪除idif id1 in list_overlapping_yellow_polygon:list_overlapping_yellow_polygon.remove(id1)if id1 in list_overlapping_blue_polygon:list_overlapping_blue_polygon.remove(id1)list_overlapping_all.clear()# 清空l(shuí)istlist_bboxs.clear()else:# 如果圖像中沒(méi)有任何的bbox，則清空l(shuí)istlist_overlapping_blue_polygon.clear()list_overlapping_yellow_polygon.clear()# 輸出計(jì)數(shù)信息text_draw = 'DOWN: ' + str(down_count) + \' , UP: ' + str(up_count)output_image_frame = cv2.putText(img=output_image_frame, text=text_draw,org=draw_text_postion,fontFace=font_draw_number,fontScale=0.75, color=(0, 0, 255), thickness=2)cv2.imshow('Counting Demo', output_image_frame)cv2.waitKey(1)capture.release()cv2.destroyAllWindows()

若需要更改模型，只需要更改objdetector.py下面的給出的部分：

OBJ_LIST = ['person', 'car', 'bus', 'truck'] DETECTOR_PATH = 'weights/yolov5m.pt'

---

總結(jié)

本篇文章給出了基于yolov5與Deep Sort的流量統(tǒng)計(jì)與軌跡跟蹤的實(shí)例，在項(xiàng)目中有著實(shí)際的應(yīng)用場(chǎng)景。
下面給出源碼地址，歡迎star：
https://github.com/JulyLi2019/yolov5-deepsort/releases/tag/V1.0，yolov5-deepsort.zip文件
如果閱讀本文對(duì)你有用，歡迎一鍵三連呀！！！
2022年4月15日09:59:53

總結(jié)

以上是生活随笔為你收集整理的基于yolov5与Deep Sort的流量统计与轨迹跟踪的全部?jī)?nèi)容，希望文章能夠幫你解決所遇到的問(wèn)題。

如果覺(jué)得生活随笔網(wǎng)站內(nèi)容還不錯(cuò)，歡迎將生活随笔推薦給好友。