Presentation_internship_rough Draft-2.pptx

INTERNSHIP PROJECT

OBJECT DETECTION SYSTEM WITH TODA: TensorFlow Object Detection API

People in Charge Internship Project

Gede Agus A.Sani

Poj Tangamchit

Student

Advisor

COMPUTER VISION TASK

R-CNN

1. Selective Search for Generating Regional Proposals 2. CNN for every regional proposals to compute features 3. Classify each region using class-specific linear-SVMs

4. Bounding box regressor

TENSORFLOW OBJECT DETECTION API (TODA)

is an open source Framework built on top of TensorFlow that makes it easy to construct, train, and, deploy object

detection models.

GRAPH IN TENSORFLOW

TRAINED MODEL TensorFlow provide some model, trained on COCO dataset, Kitti dataset, and Open Image dataset.

3 RESULT

TODA IMPLEMENTATION IMPORT MODEL PREPARATION LOAD MODEL & LABEL MAP IMAGE PREPARATION DETECTION & VISUALIZATION

LM

VIS

TODA IMPLEMENTATION

MODEL COMPARISON

A. ssd_mobilenet_v1_coco, B. ssd_mobilenet_v2_coco, C. ssd_inception_v2_coco, D. ssdlite_mobilenet_v2_coco,

E. faster_rcnn_inception_v2, F. faster_rcnn_resnet50_coco

A

B

D

E

C

F

A

B

D

E

C

F

A

B

D

E

C

F

A

B

C

27.35 seconds

37.16 seconds

54.11 seconds

D

E

27.90 seconds

48.74 seconds

F

80.67 seconds

MAIN SYSTEM

START Import Requirements

LM utils

VIS utils

Prepare Model

Model Exist ?

No

Yes Extract Model

Load model & Label map to Memory

Cont

CONT

CONT

Capture Video into Frames

Feed frames to model

Is person in text and isplay?

Feed the result to vis utils Append the label to 2 text file

Evaluate second text file & isplay

Evaluate second text file & isplay

Play the Sound

No

Evaluate first text file & isplay

Is count enough ?

Yes

Write time and detection result in new document

Update isplay into False