Published October 27, 2019 © MIT

Detecting Object for Vehicle Blind Spot by M5StickV

Detect cars and people which is blinding by drivers. You need only putting M5Stack and you aren't doing dedicated construction.

BeginnerShowcase (no instructions)5 hours5,760

Detecting Object for Vehicle Blind Spot by M5StickV

Things used in this project

Hardware components

M5Stack M5StickV

Flash Memory Card, MicroSDHC Card

I bought 32GB and choose lexar. https://www.amazon.com/dp/B012PLSCQ0

Software apps and online services

M5Stack V-Training

I processed it. https://docs.m5stack.com/#/en/related_documents/v-training

Story

Introduction

Some of the car accidents have disappeared due to pedestrians and cars entering the blind spot of the driver. So, I decided make cheap system detecting vehicle and person by M5StickV and V-Training.

Requirement definition

The driver cannot take action on the system because it is assumed that the car is driven.
When a pedestrian or car enters the driver's blind spot, it notifies the driver accordingly.
No wiring to the car.

How do it?

Inferring by taking more than 35 sample images from a camera using a program provided by M5Stack and sending them to a system called V-Train can be solved detection.

If you detection vehicle and people, you must take picture vehicle and people. So, this service is useful which getting sample data.

First, get boot.zip for V-Train. The source is the following website(https://docs.m5stack.com/#/en/related_documents/v-training?id=material-training). If the M5StickV firmware is not latest, please update it according to the procedure on the following website(https://docs.m5stack.com/#/en/related_documents/v-training?id=burner-firmware).

Second, you must take picture vehicle and people. If you got any driving movie file, it's very useful for getting any sample.

OXFORD ROBOTCAR DATASET https://robotcar-dataset.robots.ox.ac.uk/datasets/
APOLLOSCAPE-carinstance http://apolloscape.auto/car_instance.html

Here, an example of the acquired picture is shown.

Example of people picture.

Example of vehicle picture.

It shows how I take picture. I played sample video on my screen and took a picture of the video.

I define class 1 is people and class 2 is vehicle. And I got picture more than 100 per class. If you need accuracy, please get more picture.

result folder

Third, you must send SD data(train and valid) V-traing system. Please zip two folder for one achive and send this form(http://v-training.m5stack.com/) with your email address. And waiting about 30 minites!

Upload form

After 30 minutes, V-Training result will be send your email address.

Cross Validation result(60%...Ummm...)

Finally, Please backup your microSD and remove any file. If you satisfaction accuracy, copy your kmodel file and boot.py which is release in here. Current boot.py is not load your kmodel file. So, please change XXXXXX.kmodel to your kmodel file's name.

Please insert your micro SD to M5Stack and boot by power button. Detection will be run!

Result

M5StickV's LED is blinking any color by changing focused object.

See also...

Adding blinking somewhere by Obniz and Node-RED(enebular)

https://www.slideshare.net/MinoruInoue2/m5stickvobnizenebular

boot.py

import image
import lcd
import sensor
import sys
import time
import KPU as kpu
from fpioa_manager import fm
from machine import UART
from Maix import GPIO

import KPU as kpu

# UART
fm.register(35, fm.fpioa.UART2_TX, force=True)
fm.register(34, fm.fpioa.UART2_RX, force=True)
uart_port = UART(UART.UART2, 115200, 8, 0, 0,\
    timeout=1000, read_buf_len=4096)

# LED
fm.register(6, fm.fpioa.GPIO0)
fm.register(9, fm.fpioa.GPIO1)
fm.register(8, fm.fpioa.GPIO2)
fm.register(7, fm.fpioa.GPIO3)
led_r = GPIO(GPIO.GPIO0, GPIO.OUT)
led_r.value(1)
led_g = GPIO(GPIO.GPIO1, GPIO.OUT)
led_g.value(1)
led_b = GPIO(GPIO.GPIO2, GPIO.OUT)
led_b.value(1)
led_w = GPIO(GPIO.GPIO3, GPIO.OUT)
led_w.value(1)


lcd.init()
lcd.rotation(2)

task = kpu.load("/sd/XXXXXX.kmodel")

labels=["1","2"] #You can check the numbers here to real names.

sensor.reset()
sensor.set_pixformat(sensor.RGB565)
sensor.set_framesize(sensor.QVGA)
sensor.set_windowing((224, 224))
sensor.run(1)

lcd.clear()


while(True):
    img = sensor.snapshot()
    fmap = kpu.forward(task, img)
    plist=fmap[:]
    pmax=max(plist)
    max_index=plist.index(pmax)
    a = lcd.display(img)
    if pmax > 0.95:
        lcd.draw_string(40, 60, "Accu:%.2f Type:%s"%(pmax, labels[max_index].strip()))
        if labels[max_index] == "1":
            led_r.value(0)
            led_g.value(1)
            led_b.value(1)
            led_w.value(1)
        elif labels[max_index] == "2":
            led_r.value(1)
            led_g.value(0)
            led_b.value(1)
            led_w.value(1)
        uart_port.write(bytearray("LEDON"))
        time.sleep_ms(10)
    else:
        led_r.value(1)
        led_g.value(1)
        led_b.value(0)
        led_w.value(0)
        uart_port.write(bytearray("LEDOFF"))
        time.sleep_ms(10)

a = kpu.deinit(task)

Credits

Minoru Inoue

1 project • 3 followers

Detecting Object for Vehicle Blind Spot by M5StickV

Things used in this project

Hardware components

Software apps and online services

Story

Introduction

Requirement definition

How do it?

Result

Code

boot.py

Credits

Minoru Inoue

Comments

Embed the widget on your own site

Detecting Object for Vehicle Blind Spot by M5StickV

Detecting Object for Vehicle Blind Spot by M5StickV

Things used in this project

Hardware components

Software apps and online services

Story

Introduction

Requirement definition

How do it?

Result

Code

boot.py

Credits

Minoru Inoue

Comments

Related channels and tags