Google Street View object detection.
generated on 29/Jan/2018. This demo uses the EAST text detection model and the Faster RCNN with Inception Resnet v2 object detection model trained on objects in Microsoft's COCO Dataset. We used , on images automatically scrapped from . instructions:
  • click on dots on map to see Street View images and detections.
  • click on object names to see their spatial distribution.
Click here to see some basic analyis of appearance correlations of detected objects. 		Street view image sampling: downtown Madrid, an area of aprox 1km X 700m, every street corner and random street midpoints. At each sampling point images were grabbed facing front, left and right as the algorithm automatically walked through sreet view. This represents a total of 629 locations (black dots on map) and 1887 images.

Objects detected (click to see on map) in parenthesis the number of total dectections.
texts (6911) person (5634) car (2876) motorcycle (713) chair (420) truck (311) bicycle (171) potted plant (170) bus (158) handbag (143) backpack (101) umbrella (100) traffic light (95) stop sign (85) bench (59) dining table (40) sports ball (29) suitcase (27) clock (15) fire hydrant (12) refrigerator (11) bottle (11) book (8) vase (8) tv (8) dog (7) train (7) skateboard (7) parking meter (7) boat (5) skis (4) couch (3) cat (3) cell phone (3) frisbee (3) baseball glove (2) toilet (2) cup (2) horse (2) bowl (2) bird (1) banana (1) wine glass (1) keyboard (1) kite (1) oven (1) remote (1) snowboard (1) surfboard (1) teddy bear (1) tennis racket (1) tie (1) laptop (1)
Challenge: find the teddy bear and the tie

Joint behaviours

People and cars
Observe that 583 locations contained people, out of which about 130 locations or 22% had no cars. However, out of the 478 locations containing cars, only 5% had no people. This might be evidence of public spaces for pedestrians with no cars but very few locations ONLY for cars

Backpacks and people
Backpacks mostly appear around people as only one location detected a backpack without people. However, few people carry backpacks as only around 15% of scenes with people contained backpacks (less than 100 out of 583).

Plants and trucks
Potted plants might be an indicator of residential neighbourhoods, which seem quite disjoint from trucks, which might signal commercial areas. Observe that most locations with trucks show no plants and inversely.